file_name
stringlengths 7
27
| editorial
stringlengths 7
3.02k
| journal
stringclasses 3
values | abstracts
stringlengths 30
6.91k
| img
imagewidth (px) 505
1.25k
|
|---|---|---|---|---|
1_1.png | ONLINE COVER A Hop, Skip, and a JUMP! Shown is the integrated robotic platform Salto. Optimal jumping requires the efficient use of instantaneous power from the combined and modulated motion of muscle and tendon, often exceeding what can be achieved by muscle alone. Drawing inspiration from animal agility studies, Haldane et al. developed a leg mechanism with improved power modulation to create a robot with exceptional single jump and sequential jumping capabilities. Science Robotics is launching to promote the communication of fundamental research, general principles, original developments, and new applications of robotics. [CREDIT: D. HALDANE] | science | Several arboreal mammals have the ability to rapidly and repeatedly jumpvertical distances of 2 m, starting from rest. We characterize thisperformance by a metric we call vertical jumping agility. Through basickinetic relations, we show that this agility metric is fundamentallyconstrained by available actuator power. Although rapid high jumping is animportant performance characteristic, the ability to control forces duringstance also appears critical for sophisticated behaviors. The animal with thehighest vertical jumping agility, the galago ( Galago senegalensis ), isknown to use a power-modulating strategy to obtain higher peak power than thatof muscle alone. Few previous robots have used series-elastic power modulation(achieved by combining series-elastic actuation with variable mechanicaladvantage), and because of motor power limits, the best current robot has avertical jumping agility of only 55% of a galago. Through use of a specializedleg mechanism designed to enhance power modulation, we constructed a jumpingrobot that achieved 78% of the vertical jumping agility of a galago. Agilerobots can explore venues of locomotion that were not previously attainable.We demonstrate this with a wall jump, where the robot leaps from the floor toa wall and then springs off the wall to reach a net height that is greaterthan that accessible by a single jump. Our results show that series-elasticpower modulation is an actuation strategy that enables a clade of verticallyagile robots.
| |
2_10.png | ONLINE COVER Biological Remora with the Bioinspired Robotic Disc. Remoras attach to a diverse array of marine animals using a highly modified dorsal fin that forms an adhesive disc. Wang et al. developed a multimaterial biomimetic disc that enables robust attachment and underwater hitchhiking to a variety of surfaces. This bioinspired adhesive technology may reduce transport and movement costs, increase mission durations for autonomous underwater vehicles, and enable underwater gripping applications. [CREDIT: LI WEN, YUEPING WANG, AND KLAUS STIEFEL] | science | Remoras of the ray-finned fish family Echeneidae have the remarkable abilityto attach to diverse marine animals using a highly modified dorsal fin thatforms an adhesive disc, which enables hitchhiking on fast-swimming hostsdespite high magnitudes of fluid shear. We present the design of abiologically analogous, multimaterial biomimetic remora disc based on detailedmorphological and kinematic investigations of the slender sharksucker (Echeneis naucrates ). We used multimaterial three-dimensional printingtechniques to fabricate the main disc structure whose stiffness spans threeorders of magnitude. To incorporate structures that mimic the functionality ofthe remora lamellae, we fabricated carbon fiber spinules (270 **** m basediameter) using laser machining techniques and attached them to softactuatorcontrolled lamellae. Our biomimetic prototype can attach to differentsurfaces and generate considerable pull-off forceup to 340 times the weightof the disc prototype. The rigid spinules and soft material overlaying thelamellae engage with the surface when rotated, just like the discs of liveremoras. The biomimetic kinematics result in significantly enhanced frictionalforces across the disc on substrates of different roughness. Using ourprototype, we have designed an underwater robot capable of strong adhesion andhitchhiking on a variety of surfaces (including smooth, rough, and compliantsurfaces, as well as shark skin). Our results demonstrate that there ispromise for the development of high-performance bioinspired robotic systemsthat may be used in a number of applications based on an understanding of theadhesive mechanisms used by remoras.
| |
2_13.png | ONLINE COVER How to be Human. Humanoids like Kengoro, developed by Asano et al., represent one of the ultimate goals of robotics: to synthesize advances from many disciplines. These robots are inherently cross-disciplinary, involving advanced locomotion and manipulation, biomechanics, artificial intelligence, machine vision, perception, learning and cognitive development, as well as behavioral studies. [CREDIT: Y. ASANO ET AL./SCIENCE ROBOTICS] | science | Many systems and mechanisms in the human body are not fully understood, suchas the principles of muscle control, the sensory nervous system that connectsthe brain and the body, learning in the brain, and the human walking motion.To address this knowledge deficit, we propose a human mimetic humanoid with anunprecedented degree of anatomical fidelity to the human musculoskeletalstructure. The fundamental concept underlying our design is to consider thehuman mechanism, which contrasts with the conventional engineering approachused in the design of existing humanoids. We believe that the proposed humanmimetic humanoid can be used to provide new opportunities in science, forinstance, to quantitatively analyze the internal data of a human body inmovement. We describe the principles and development of human mimetichumanoids, Kenshiro and Kengoro, and compare their anatomical fidelity withhumans in terms of body proportions, skeletal structures, muscle arrangement,and joint performance. To demonstrate the potential of human mimetichumanoids, Kenshiro and Kengoro performed several typical human motions.
| |
2_2.png | ONLINE COVER Step This Way. A volunteer walks on a treadmill while wearing a soft exosuit designed to assist the wearer and to measure gait mechanics Quinlivan et al. track the energy expended during walking with and without assistance from the device. Force is applied locally to the ankle to closely approximate normal ankle motion and indirectly applied to hip motion through the suit's fabrics. Net metabolic rates showed progressive decrease throughout the entire range of exosuit assistance. [CREDIT: TAYLOR GREENBERG GOLDY] | science | When defining requirements for any wearable robot for walking assistance, itis important to maximize the users metabolic benefit resulting from theexosuit assistance while limiting the metabolic penalty of carrying thesystems mass. Thus, the aim of this study was to isolate and characterize therelationship between assistance magnitude and the metabolic cost of walkingwhile also examining changes to the wearers underlying gait mechanics. Thestudy was performed with a tethered multiarticular soft exosuit during normalwalking, where assistance was directly applied at the ankle joint andindirectly at the hip due to a textile architecture. The exosuit controllerwas designed such that the delivered torque profile at the ankle jointapproximated that of the biological torque during normal walking. Sevenparticipants walked on a treadmill at 1.5 meters per second under oneunpowered and four powered conditions, where the peak moment applied at theankle joint was varied from about 10 to 38% of biological ankle moment(equivalent to an applied force of 18.7 to 75.0% of body weight). Resultsshowed that, with increasing exosuit assistance, net metabolic ratecontinually decreased within the tested range. When maximum assistance wasapplied, the metabolic rate of walking was reduced by 22.83 3.17% relativeto the powered-off condition (mean SEM).
| |
2_3.png | ONLINE COVER Some Batty Ideas Take Flight. Bat Bot (shown) is able to imitate several flight maneuvers of bats, such as bank turns and diving flight. Such agile flight is made possible by highly malleable bones and skin in bat wings. Ramezani et al. identified and implemented the most important bat wing joints by means of a series of mechanical constraints. They then designed feedback control for their flapping wing platform and covered the structure in a flexible silicon membrane. This biomimetic robot may also shed light on the role of bat legs to modulate flight pitch. [CREDIT: ALIREZA RAMEZANI] | science | Bats have long captured the imaginations of scientists and engineers withtheir unrivaled agility and maneuvering characteristics, achieved byfunctionally versatile dynamic wing conformations as well as more than 40active and passive joints on the wings. Wing flexibility and complex wingkinematics not only bring a unique perspective to research in biology andaerial robotics but also pose substantial technological challenges for robotmodeling, design, and control. We have created a fully self-contained,autonomous flying robot that weighs 93 grams, called Bat Bot (B2), to mimicsuch morphological properties of bat wings. Instead of using a large number ofdistributed control actuators, we implement highly stretchable silicone-basedmembrane wings that are controlled at a reduced number of dominant wing jointsto best match the morphological characteristics of bat flight. First, thedominant degrees of freedom (DOFs) in the bat flight mechanism are identifiedand incorporated in B2s design by means of a series of mechanicalconstraints. These biologically meaningful DOFs include asynchronous andmediolateral movements of the armwings and dorsoventral movements of the legs.Second, the continuous surface and elastic properties of bat skin under wingmorphing are realized by an ultrathin (56 micrometers) membranous skin thatcovers the skeleton of the morphing wings. We have successfully achievedautonomous flight of B2 using a series of virtual constraints to control thearticulated, morphing wings.
| |
2_8.png | ONLINE COVER From the Tip Leveraging the flexibility that is possible when using soft materials, Hawkes et al. used pneumatics to extend the tip of a robot over an extended distance. They inflated chambers on only one side of the robot to make it turn. An onboard camera connected to a feedback loop allowed the robot to track toward a distant light source. [CREDIT: LINDA A. CICERO/STANFORD NEWS SERVICE] | science | Across kingdoms and length scales, certain cells and organisms navigate theirenvironments not through locomotion but through growth. This pattern ofmovement is found in fungal hyphae, developing neurons, and trailing plants,and is characterized by extension from the tip of the body, length change ofhundreds of percent, and active control of growth direction. This results inthe abilities to move through tightly constrained environments and form usefulthree-dimensional structures from the body. We report a class of softpneumatic robot that is capable of a basic form of this behavior, growingsubstantially in length from the tip while actively controlling directionusing onboard sensing of environmental stimuli; further, the peak rate oflengthening is comparable to rates of animal and robot locomotion. This isenabled by two principles: Pressurization of an inverted thin-walled vesselallows rapid and substantial lengthening of the tip of the robot body, andcontrolled asymmetric lengthening of the tip allows directional control.Further, we demonstrate the abilities to lengthen through constrainedenvironments by exploiting passive deformations and form three-dimensionalstructures by lengthening the body of the robot along a path. Our study helpslay the foundation for engineered systems that grow to navigate theenvironment.
| |
2_9.png | ONLINE COVER Malleable Machines. Advances in materials science often improve machines as well. Terryn et al. construct their robotics using self-healing polymers that exploit Diels-Alders reactions. They demonstrate a soft gripper, a soft hand, and artificial muscles that can be healed with mild heat treatment. The polymers also have the advantage of being recyclable. [CREDIT: SEPPE TERRYN] | science | Inspired by the compliance found in many organisms, soft robots are madealmost entirely out of flexible, soft material, making them suitable forapplications in uncertain, dynamic task environments, including safe human-robot interactions. Their intrinsic compliance absorbs shocks and protectsthem against mechanical impacts. However, the soft materials used for theirconstruction are highly susceptible to damage, such as cuts and perforationscaused by sharp objects present in the uncontrolled and unpredictableenvironments they operate in. In this research, we propose to construct softrobotics entirely out of self-healing elastomers. On the basis of healingcapacities found in nature, these polymers are given the ability to healmicroscopic and macroscopic damage. Diels-Alder polymers, beingthermoreversible covalent networks, were used to develop three applications ofself-healing soft pneumatic actuators (a soft gripper, a soft hand, andartificial muscles). Soft pneumatic actuators commonly experience perforationsand leaks due to excessive pressures or wear during operation. All threeprototypes were designed using finite element modeling and mechanicallycharacterized. The manufacturing method of the actuators exploits the self-healing behavior of the materials, which can be recycled. Realisticmacroscopic damage could be healed entirely using a mild heat treatment. Atthe location of the scar, no weak spots were created, and the full performanceof the actuators was nearly completely recovered after healing.
| |
3_15.png | ONLINE COVER Making Skins Crawl. Inspired by snake skins and the art of kirigami, Rafsanjani et al. designed flat plastic sheets that could be transformed into three-dimensional textured surfaces. When the kirigami sheets were wrapped around a tubular soft actuator, buckle-induced directional friction enabled the robotic system to efficiently crawl. [CREDIT: AHMAD RAFSANJANI/HARVARD UNIVERSITY] | science | Bioinspired soft machines made of highly deformable materials are enabling avariety of innovative applications, yet their locomotion typically requiresseveral actuators that are independently activated. We harnessed kirigamiprinciples to significantly enhance the crawling capability of a softactuator. We designed highly stretchable kirigami surfaces in which mechanicalinstabilities induce a transformation from flat sheets to 3D-textured surfacesakin to the scaled skin of snakes. First, we showed that this transformationwas accompanied by a dramatic change in the frictional properties of thesurfaces. Then, we demonstrated that, when wrapped around an extending softactuator, the buckling-induced directional frictional properties of thesesurfaces enabled the system to efficiently crawl.
| |
3_16.png | ONLINE COVER Lightweight Length. Unmanned aerial vehicles (UAVs) can move in areas that are too confined or too dangerous for humans, but many tools are too heavy or bulky to be carried by UAVs. Kim et al. developed an origami-inspired robotic arm that extends out and locks in place when needed. A gripper or other device can be added to the end of the arm, and the arm folds flat for transport or storage. [CREDIT: JAMIE JUNG-RYUL SONG/SNU SOFT ROBOTICS RESEARCH CENTER] | science | A foldable arm is one of the practical applications of folding. It can helpmobile robots and unmanned aerial vehicles (UAVs) overcome access issues byallowing them to reach into confined spaces. The origami-inspired designenables a foldable structure to be lightweight, compact, and scalable whilemaintaining its kinematic behavior. However, the lack of structural stiffnesshas been a major limitation in the practical use of origami-inspired designs.Resolving this obstacle without losing the inherent advantages of origami is achallenge. We propose a solution by implementing a simple stiffening mechanismthat uses an origami principle of perpendicular folding. The simplicity of thestiffening mechanism enables an actuation system to drive shape and stiffnesschanges with only a single electric motor. Our results show that this designwas effective for a foldable arm and allowed a UAV to perform a variety oftasks in a confined space.
| |
3_18.png | ONLINE COVER Antagonistic Actuation. Engineers have used natural muscle tissues to move robotic devices, but the tissues were limited by spontaneous shrinkage, small ranges of movement, and short lifetimes. Morimoto et al. engineered muscle tissues by stacking thin hydrogel sheets containing myoblasts, then anchored an antagonistic pair of the tissues to a flexible robotic skeleton. Actuated by electrically stimulated engineered tissues, the device achieved a joint rotation of 90, a range of motion comparable to that of a human finger. The muscle tissues remained capable of actuation for about one week. [CREDIT: SHOJI TAKEUCHI RESEARCH GROUP/UNIVERSITY OF TOKYO] | science | Biohybrid robots are attracting attention as promising candidates to enhancerobot applicability to studies on biological designs and in vitro constructionof biological dynamic systems. Rapid progress in biohybrid robots withskeletal muscle tissues formed on a flexible substrate has enabled varioustypes of locomotion powered by muscle tissue. However, it has been difficultto achieve high levels of both large and long-term actuations of the skeletalmuscle tissues because of their spontaneous shrinkage through the course ofthe tissue culture. To overcome this limitation, we adapted the concept ofbiological systems and developed a biohybrid robot actuated by an antagonisticpair of skeletal muscle tissues. Our robot achieved large actuation (90 ofrotation of a joint) by selective contractions of the skeletal muscle tissuesand a long lifetime (1 week) by balancing tensions of the antagonistictissues to prevent the spontaneous shrinkage. As a demonstration, we showedthat our biohybrid robots allowed a pick-and-place manipulation of objects.This research may provide a platform to exceed the limitations of design inconventional biohybrid robots and replicate various lifelike movements.
| |
3_22.png | ONLINE COVER Powerful and Precise. Conventional motor technologies are heavy, lack flexibility, and do not retain efficiency or power density at small scales. Kedzierski and Holihan developed linear and rotational microhydraulic actuators that have high power density and efficiency and that improved when scaled to smaller dimensions. Such lightweight, flexible, and accurate actuators may find many applications in soft robotics and biomedicine. [CREDIT: GLEN COOPER/MIT LINCOLN LABORATORY] | science | Microhydraulic actuators offer a new way to convert electrical power tomechanical power on a microscale with an unmatched combination of powerdensity and efficiency. Actuators work by combining surface tension forcecontributions from a large number of droplets distorted by electrowettingelectrodes. This paper reports on the behavior of microgram-scale linear androtational microhydraulic actuators with output force/weight ratios of 5500,cycle frequencies of 4 kilohertz, <1-micrometer movement precision, andaccelerations of 3 kilometers/second2. The power density and the efficiency ofthe actuators were characterized by simultaneously measuring the mechanicalwork performed and the electrical power applied. Maximum output power densitywas 0.93 kilowatt/kilogram, comparable with the best electric motors. Atmaximum power, the actuator was 60% efficient, but efficiencies were as highas 83% at lower power. Rotational actuators demonstrated a torque density of79 newton meters/kilogram, substantially more than electric motors ofcomparable diameter. Scaling the droplet pitch from 100 to 48 micrometersincreased power density from 0.27 to 0.93 kilowatt/kilogram, validating thequadratic scaling of actuator power.
| |
3_23.png | ONLINE COVER Tracking Progress. Recent advances in animal tags have improved remote tracking of larger species; however, this technology is not suitable for up to 70% of bird and mammal species. Efforts to automate small-animal tracking have had limited success. Cliff et al. developed an approach for tracking radio-tagged wild swift parrots by using an autonomous, lightweight aerial robot equipped with a miniaturized sensor and estimation algorithms. The robotic system was able to estimate the location of a tagged bird in its winter range as accurately and quickly as an experienced human tracker. [CREDIT: AUSTRALIAN NATIONAL UNIVERSITY] | science | Understanding animal movements that underpin ecosystem processes isfundamental to ecology. Recent advances in animal tags have increased theability to remotely locate larger species; however, this technology is notsuitable for up to 70% of the worlds bird and mammal species. The mostwidespread technique for tracking small animals is to manually locate low-power radio transmitters from the ground with handheld equipment. Despite thislabor-intensive technique being used for decades, efforts to reduce orautomate this process have had limited success. Here, we present an approachfor tracking small radio-tagged animals by using an autonomous and lightweightaerial robot. We present experimental results where we used the robot tolocate critically endangered swift parrots ( Lathamus discolor ) withintheir winter range. The system combines a miniaturized sensor with newlydeveloped estimation algorithms to yield unambiguous bearing- and range-basedmeasurements with associated measures of uncertainty. We incorporated thesemeasurements into Bayesian data fusion and information-based planningalgorithms to control the position of the robot as it collected data. Wereport estimated positions that lie within about 50 meters of the truepositions of the birds on average, which are sufficiently accurate forrecapture or observation. Further, in comparison with experienced humantrackers from locations where the signal was detectable, the robot produced acorrect estimate as fast or faster than the human. These results providevalidation of robotic systems for wildlife radio telemetry and suggest a wayfor widespread use as human-assistive or autonomous devices.
| |
4_26.png | ONLINE COVER LearningBeyond Imitation. Robot learning, rooted in machine learning and human-robot interaction, is moving beyond simple kinematic planning, environment interaction, and elementary behavioral imitation from human demonstrators. For example, universal picking, or reliable robot grasping of a diverse range of objects from heaps, is a challenging goal. Mahler et al. trained Dex-Net 4.0 on a synthetic dataset including images, grasps, and rewards. The resulting policy allowed a robot with two grippers to consistently and reliably clear bins containing up to 25 novel objects. The combination of learning from synthetic data and applying to real-world situations may improve e-commerce, manufacturing, inspection, and home service robots. [CREDIT: JEFF MAHLER, STEPHEN MCKINLEY, KEN GOLDBERG/UNIVERSITY OF CALIFORNIA, BERKELEY; PHOTO: ADRIEL OLMOS] | science | Universal picking (UP), or reliable robot grasping of a diverse range of novelobjects from heaps, is a grand challenge for e-commerce order fulfillment,manufacturing, inspection, and home service robots. Optimizing the rate,reliability, and range of UP is difficult due to inherent uncertainty insensing, control, and contact physics. This paper explores ambidextrousrobot grasping, where two or more heterogeneous grippers are used. We presentDexterity Network (Dex-Net) 4.0, a substantial extension to previous versionsof Dex-Net that learns policies for a given set of grippers by training onsynthetic datasets using domain randomization with analytic models of physicsand geometry. We train policies for a parallel-jaw and a vacuum-based suctioncup gripper on 5 million synthetic depth images, grasps, and rewards generatedfrom heaps of three-dimensional objects. On a physical robot with twogrippers, the Dex-Net 4.0 policy consistently clears bins of up to 25 novelobjects with reliability greater than 95% at a rate of more than 300 meanpicks per hour.
| |
4_27.png | ONLINE COVER Adaptive Robotic Sampling. The coastal waters near Runde Island, Norway, are affected by strong winds, ocean currents, and freshwater runoff from land. Fossum et al. used a light autonomous underwater vehicle (LAUV) to survey the edges of predefined volumes, and the resulting data allowed the robot to identify interior areas with high concentrations of subsurface chlorophyll a for additional, detailed sampling. LAUV results were confirmed with data from remote sensing and shipboard samples. The combination of real-time data analysis and accurate, adaptive robotic sampling may improve our understanding of marine food webs and their dynamic, heterogeneous environments. [CREDIT: GUNHILD ELISABETH BERGET/NORWEGIAN UNIVERSITY OF SCIENCE AND TECHNOLOGY] | science | Currents, wind, bathymetry, and freshwater runoff are some of the factors thatmake coastal waters heterogeneous, patchy, and scientificallyinterestingwhere it is challenging to resolve the spatiotemporal variationwithin the water column. We present methods and results from field experimentsusing an autonomous underwater vehicle (AUV) with embedded algorithms thatfocus sampling on features in three dimensions. This was achieved by combiningGaussian process (GP) modeling with onboard robotic autonomy, allowingvolumetric measurements to be made at fine scales. Special focus was given tothe patchiness of phytoplankton biomass, measured as chlorophyll a (Chla), animportant factor for understanding biogeochemical processes, such as primaryproductivity, in the coastal ocean. During multiple field tests in Runde,Norway, the method was successfully used to identify, map, and track thesubsurface chlorophyll a maxima (SCM). Results show that the algorithm wasable to estimate the SCM volumetrically, enabling the AUV to track the maximumconcentration depth within the volume. These data were subsequently verifiedand supplemented with remote sensing, time series from a buoy and ship-basedmeasurements from a fast repetition rate fluorometer (FRRf), particle imagingsystems, as well as discrete water samples, covering both the large and smallscales of the microbial community shaped by coastal dynamics. By bringingtogether diverse methods from statistics, autonomous control, imaging, andoceanography, the work offers an interdisciplinary perspective in roboticobservation of our changing oceans.
| |
4_28.png | ONLINE COVER Stable Approaches to Landing. Unmanned aerial vehicles (UAVs) are affordable tools for many tasks, such as surveillance and delivery, but constant motion quickly drains their batteries. Flying animals are able to attach to or perch on top of structures, conserving energy between tasks. Hang et al. designed modular, actuating landing gear that attached to a commercial, off-the-shelf quadrotor UAV. The UAV was able to approach objects without collisions and actuate the landing modules to perch on and/or grasp objects without damage. It also maintained its position once stabilized, allowing continued surveillance with low energy use. [CREDIT: KAIYU HANG/YALE UNIVERSITY] | science | Perching helps small unmanned aerial vehicles (UAVs) extend their time ofoperation by saving battery power. However, most strategies for UAV perchingrequire complex maneuvering and rely on specific structures, such as roughwalls for attaching or tree branches for grasping. Many strategies to perchingneglect the UAVs mission such that saving battery power interrupts themission. We suggest enabling UAVs with the capability of making andstabilizing contacts with the environment, which will allow the UAV to consumeless energy while retaining its altitude, in addition to the perchingcapability that has been proposed before. This new capability is termedresting. For this, we propose a modularized and actuated landing gearframework that allows stabilizing the UAV on a wide range of differentstructures by perching and resting. Modularization allows our framework toadapt to specific structures for resting through rapid prototyping withadditive manufacturing. Actuation allows switching between different modes ofperching and resting during flight and additionally enables perching bygrasping. Our results show that this framework can be used to perform UAVperching and resting on a set of common structures, such as street lights andedges or corners of buildings. We show that the design is effective inreducing power consumption, promotes increased pose stability, and preserveslarge vision ranges while perching or resting at heights. In addition, wediscuss the potential applications facilitated by our design, as well as thepotential issues to be addressed for deployment in practice.
| |
4_32.png | ONLINE COVER ACES Architecture. Electronic skins (e-skins) can facilitate robotic interactions with objects and environments, but equipping e-skins with enough sensors for precise and rapid data processing has been difficult. Lee et al. developed the Asynchronously Coded Electronic Skin (ACES) system, which mimics biological action potentials and mechanoreceptors to transmit thermotactile information from sensors in e-skin to a single conductor without complex wiring or signaling delays. ACES detected heat and pressure with high temporal resolution, even across dozens of sensors; was not susceptible to background electromagnetic interference; and continued to operate after mechanical damage to the e-skin. [CREDIT: BENJAMIN C. K. TEE/NATIONAL UNIVERSITY OF SINGAPORE] | science | The human sense of touch is essential for dexterous tool usage, spatialawareness, and social communication. Equipping intelligent human-like androidsand prosthetics with electronic skinsa large array of sensors spatiallydistributed and capable of rapid somatosensory perceptionwill enable them towork collaboratively and naturally with humans to manipulate objects inunstructured living environments. Previously reported tactile-sensitiveelectronic skins largely transmit the tactile information from sensorsserially, resulting in readout latency bottlenecks and complex wiring as thenumber of sensors increases. Here, we introduce the Asynchronously CodedElectronic Skin (ACES)a neuromimetic architecture that enables simultaneoustransmission of thermotactile information while maintaining exceptionally lowreadout latencies, even with array sizes beyond 10,000 sensors. We demonstrateprototype arrays of up to 240 artificial mechanoreceptors that transmittedevents asynchronously at a constant latency of 1 ms while maintaining anultra-high temporal precision of <60 ns, thus resolving fine spatiotemporalfeatures necessary for rapid tactile perception. Our platform requires only asingle electrical conductor for signal propagation, realizing sensor arraysthat are dynamically reconfigurable and robust to damage. We anticipate thatthe ACES platform can be integrated with a wide range of skin-like sensors forartificial intelligence (AI)enhanced autonomous robots, neuroprosthetics, andneuromorphic computing hardware for dexterous object manipulation andsomatosensory perception.
| |
4_33.png | ONLINE COVER Soft Robotics Across Scales. Extensive research, from materials engineering to whole systems, is expanding the boundaries of robotics to create stretchy, flexible, touch-feedback electronics; advanced microelectromechanical systems; and shape-morphing and self-propelling devices. Umrao et al. designed Ti3C2Tx electrodes ionically cross-linked with PEDOT:PSS for high-performance air-working ionic soft actuators. The resulting actuators tolerated high bending strains, responded rapidly to electric signals, and remained durable after hours of use. They were used as artificial muscles in kinetic structures, such as this butterfly (mass of 310 grams). [CREDIT: Umrao et al./Science Robotics] | science | Existing ionic artificial muscles still require a technology breakthrough formuch faster response speed, higher bending strain, and longer durability.Here, we report an MXene artificial muscle based on ionically cross-linkedTi3C2T x with poly(3,4 ethylenedioxythiophene)-poly(styrenesulfonate),showing ultrafast rise time of within 1 s in DC responses, extremely largebending strain up to 1.37% in very low input voltage regime (0.1 to 1 V),long-term cyclic stability of 97% up to 18,000 cycles, markedly reduced phasedelay, and very broad frequency bandwidth up to 20 Hz with good structuralreliability without delamination under continuous electrical stimuli. Theseartificial muscles were successfully applied to make an origami-inspirednarcissus flower robot as a wearable brooch and dancing butterflies and leaveson a tree as a kinetic art piece. These successful demonstrations elucidatethe wide potential of MXene-based soft actuators for the next-generation softrobotic devices including wearable electronics and kinetic art pieces.
| |
4_34.png | ONLINE COVER Sensors in Soft 3D Scaffolds. The sensory systems of living organisms are able to detect myriad stimuli across their bodies. In contrast, robotic sensors are typically limited to specific areas of a robot and perceive one type of stimulus. Xu et al. created a platform, called optical lace, to construct arbitrary 3D grids of soft, stretchable light guides for spatially discontinuous deformation sensing. Devices composed of optical lace reliably and repeatedly sensed external and internal deformations. Optical lace may be expanded to cover more area and/or modified to improve resolution or to detect other stimuli, such as temperature. [CREDIT: ROBERT SHEPHERD/CORNELL UNIVERSITY] | science | Whereas vision dominates sensing in robots, animals with limited vision deftlynavigate their environment using other forms of perception, such as touch.Efforts have been made to apply artificial skins with tactile sensing torobots for similarly sophisticated mobile and manipulative skills. The abilityto functionally mimic the afferent sensory neural network, required fordistributed sensing and communication networks throughout the body, is stillmissing. This limitation is partially due to the lack of cointegration of themechanosensors in the body of the robot. Here, lacings of stretchable opticalfibers distributed throughout 3D-printed elastomer frameworks created acointegrated body, sensing, and communication network. This soft, functionalstructure could localize deformation with submillimeter positional accuracy(error of 0.71 millimeter) and sub-Newton force resolution (0.3 newton).
| |
4_35.png | ONLINE COVER Providing a SPARC. Children respond well to social robots, interacting with them more than with software on a tablet or computer. Senft et al. used a machine learning technique called SPARC (supervised progressively autonomous robot competencies) to enable social robots to learn from a (human) teacher how to coach children playing an educational game. Initially, the robots did not act unless remotely controlled by the teacher; as the teacher selected actions, SPARC learned which behaviors were appropriate for particular situations. The technique may allow domain experts to quickly train social robots to autonomously choose safe actions without requiring specialized programming skills. [CREDIT: PIKUL NOOROD/SHUTTERSTOCK.COM] | science | Striking the right balance between robot autonomy and human control is a corechallenge in social robotics, in both technical and ethical terms. On the onehand, extended robot autonomy offers the potential for increased humanproductivity and for the off-loading of physical and cognitive tasks. On theother hand, making the most of human technical and social expertise, as wellas maintaining accountability, is highly desirable. This is particularlyrelevant in domains such as medical therapy and education, where social robotshold substantial promise, but where there is a high cost to poorly performingautonomous systems, compounded by ethical concerns. We present a field studyin which we evaluate SPARC (supervised progressively autonomous robotcompetencies), an innovative approach addressing this challenge whereby arobot progressively learns appropriate autonomous behavior from in situ humandemonstrations and guidance. Using online machine learning techniques, wedemonstrate that the robot could effectively acquire legible and congruentsocial policies in a high-dimensional child-tutoring situation needing only alimited number of demonstrations while preserving human supervision wheneverdesirable. By exploiting human expertise, our technique enables rapid learningof autonomous social and domain-specific policies in complex andnondeterministic environments. Last, we underline the generic properties ofSPARC and discuss how this paradigm is relevant to a broad range of difficulthuman-robot interaction scenarios.
| |
4_36.png | ONLINE COVER From flat to functional. Origami is a popular inspiration for design of inexpensive, easily assembled robotic systems. Kim et al. took inspiration from Eurypharynx pelecanoides (commonly called the pelican eel), which has the ability to rapidly expand its mouth to capture prey. They designed soft robots made of silicone rubber that can unfold and stretch simultaneously. One origami robot could mimic the mouth movement of E. pelecanoides. Modifications in the fold designs and the stiffness of components allowed programmable designs and shapes, including crawlers and grippers that can be stored flat. [CREDIT: JAMIE SONG/SNU SRRC] | science | Nature demonstrates adaptive and extreme shape morphing via unique patterns ofmovement. Many of them have been explained by monolithic shape-changingmechanisms, such as chemical swelling, skin stretching, origami/kirigamimorphing, or geometric eversion, that were successfully mimicked in artificialanalogs. However, there still remains an unexplored regime of natural morphingthat cannot be reproduced in artificial systems by a single-mode morphingmechanism. One example is the dual-mode morphing of Eurypharynxpelecanoides (commonly known as the pelican eel), which first unfolds andthen inflates its mouth to maximize the probability of engulfing the prey.Here, we introduce pelican eelinspired dual-morphing architectures thatembody quasi-sequential behaviors of origami unfolding and skin stretching inresponse to fluid pressure. In the proposed system, fluid paths were enclosedand guided by a set of entirely stretchable origami units that imitate themorphing principle of the pelican eels stretchable and foldable frames. Thisgeometric and elastomeric design of fluid networks, in which fluid pressureacts in the direction that the whole body deploys first, resulted in a quasi-sequential dual-morphing response. To verify the effectiveness of our designrule, we built an artificial creature mimicking a pelican eel and reproducedbiomimetic dual-morphing behavior. By compositing the basic dual-morphing unitcells into conventional origami frames, we demonstrated architectures of softmachines that exhibit deployment-combined adaptive gripping, crawling, andlarge range of underwater motion. This design principle may provide guidancefor designing bioinspired, adaptive, and extreme shape-morphing systems.
| |
4_37.png | ONLINE COVER Explainable Artificial Intelligence (XAI). Recent successes in machine learning have lead to new AI applications that offer important benefits. However, many of these systems are not able to explain their autonomous decisions and actions to humans, a critical ability in areas such as defense, medicine, and finance. Edmonds et al. used a robotic system that could learn to open bottles with various safety mechanisms. The robot then demonstrated this ability to people with various explanations, and the authors noted which explanations best fostered trust. Li et al. developed an interpretable, safe approach to reinforcement learning that was then applied to robots cooking and serving hot dogs. [CREDIT: IMAGINIMA/ISTOCK.COM] | science | The ability to provide comprehensive explanations of chosen actions is ahallmark of intelligence. Lack of this ability impedes the general acceptanceof AI and robot systems in critical tasks. This paper examines what forms ofexplanations best foster human trust in machines and proposes a framework inwhich explanations are generated from both functional and mechanisticperspectives. The robot system learns from human demonstrations to openmedicine bottles using (i) an embodied haptic prediction model to extractknowledge from sensory feedback, (ii) a stochastic grammar model induced tocapture the compositional structure of a multistep task, and (iii) an improvedEarley parsing algorithm to jointly leverage both the haptic and grammarmodels. The robot system not only shows the ability to learn from humandemonstrators but also succeeds in opening new, unseen bottles. Usingdifferent forms of explanations generated by the robot system, we conducted apsychological experiment to examine what forms of explanations best fosterhuman trust in the robot. We found that comprehensive and real-timevisualizations of the robots internal decisions were more effective inpromoting human trust than explanations based on summary text descriptions. Inaddition, forms of explanation that are best suited to foster trust do notnecessarily correspond to the model components contributing to the best taskperformance. This divergence shows a need for the robotics community tointegrate model components to enhance both task execution and human trust inmachines.
Growing interest in reinforcement learning approaches to robotic planning andcontrol raises concerns of predictability and safety of robot behaviorsrealized solely through learned control policies. In addition, formallydefining reward functions for complex tasks is challenging, and faulty rewardsare prone to exploitation by the learning agent. Here, we propose a formalmethods approach to reinforcement learning that (i) provides a formalspecification language that integrates high-level, rich, task specificationswith a priori, domain-specific knowledge; (ii) makes the reward generationprocess easily interpretable; (iii) guides the policy generation processaccording to the specification; and (iv) guarantees the satisfaction of the(critical) safety component of the specification. The main ingredients of ourcomputational framework are a predicate temporal logic specifically tailoredfor robotic tasks and an automaton-guided, safe reinforcement learningalgorithm based on control barrier functions. Although the proposed frameworkis quite general, we motivate it and illustrate it experimentally for arobotic cooking task, in which two manipulators worked together to make hotdogs.
| |
5_38.png | ONLINE COVER Winging It. Bioinspired robotics translates fundamental biological principles into engineering design rules to create robots that perform like natural systems. Inspired by how birds demonstrate agile and efficient flight through wing morphing, Chang et al. studied the skeletal and feather kinematics of the common pigeon to develop a soft biohybrid aerial robot, called PigeonBot, with underactuated feathers that can steer by morphing its wings with wrist and finger joints. This month's cover is a composite image showing the range of PigeonBot's wing morphingthe wings of this biohybrid system are composed of real pigeon feathers. [CREDIT: ERIC CHANG/STANFORD UNIVERSITY] | science | Since the Wright Flyer, engineers have strived to develop flying machines withmorphing wings that can control flight as deftly as birds. Birds morph theirwing planform parameters simultaneouslyincluding sweep, span, and areain away that has proven to be particularly challenging to embody robotically.Previous solutions have primarily centered around the classical aerospaceparadigm of controlling every degree of freedom to ensure predictableperformance, but underperform compared with birds. To understand how birdsaccomplish wing morphing, we measured the kinematics of wing flexion andextension in common pigeons, Columba livia. The skeletal and featherkinematics show that the 20 primary and 20 secondary feathers are coordinatedvia approximately linear transfer functions controlled by wrist and fingermotion. To replicate this control principle in a robot, we developed abiohybrid morphing wing with real feathers to understand the underlying designprinciples. The outcome, PigeonBot, embodies 42 degrees of freedom thatcontrol the position of 40 elastically connected feathers via four servo-actuated wrist and finger joints. Our flight tests demonstrate that the softfeathered wings morph rapidly and robustly under aerodynamic loading. They notonly enable wing morphing but also make robot interactions safer, the wingmore robust to crashing, and the wing reparable via preening. In flighttests, we found that both asymmetric wrist and finger motion can initiate turnmaneuversevidence that birds may use their fingers to steer in flight.
| |
5_40.png | ONLINE COVER Can't Touch This. To dodge, you must first perceive. To dodge quickly, you must perceive quickly. Standard frame-based cameras operate with high latencies due to their long exposure times. Thus, these cameras operate too slowly for situations requiring fast reactions. Falanga et al. use event cameras, a bioinspired neuromorphic sensor, that can distinguish between static and dynamic objects with low latency. This month's cover shows a custom-designed drone with event cameras, using only onboard sensing and computation, dodging fast-moving objects in an outdoor setting. [CREDIT: D. FALANGA ET AL./UNIVERSITY OF ZURICH] | science | Todays autonomous drones have reaction times of tens of milliseconds, whichis not enough for navigating fast in complex dynamic environments. To safelyavoid fast moving objects, drones need low-latency sensors and algorithms. Wedeparted from state-of-the-art approaches by using event cameras, which arebioinspired sensors with reaction times of microseconds. Our approach exploitsthe temporal information contained in the event stream to distinguish betweenstatic and dynamic objects and leverages a fast strategy to generate the motorcommands necessary to avoid the approaching obstacles. Standard visionalgorithms cannot be applied to event cameras because the output of thesesensors is not images but a stream of asynchronous events that encode per-pixel intensity changes. Our resulting algorithm has an overall latency ofonly 3.5 milliseconds, which is sufficient for reliable detection andavoidance of fast-moving obstacles. We demonstrate the effectiveness of ourapproach on an autonomous quadrotor using only onboard sensing andcomputation. Our drone was capable of avoiding multiple obstacles of differentsizes and shapes, at relative speeds up to 10 meters/second, both indoors andoutdoors.
| |
5_41.png | ONLINE COVER Nature's Origami. The wings of a ladybird beetle can sustain large aerodynamic forces during flight, and, when folded, these wings are deployable within 100 milliseconds. The self-locking and elastic energy storage capabilities of these wings are enabled by a tape-spring shaped vein in the wing frame. Baek et al. replicate the curved geometry of these veins to create origami structures that can self-lock and deploy within 116 milliseconds. The researchers validate their design by demonstrating a jumping robot and a jump-gliding robot with enhanced kinetic and static behavior. This month's cover shows how a ladybird beetle deploys its wings to take flight. [CREDIT: MYN/Marc Pihet/Minden Pictures] | science | Origami can enable structures that are compact and lightweight. The facets ofan origami structure in traditional designs, however, are essentiallynondeformable rigid plates. Therefore, implementing energy storage and robustself-locking in these structures can be challenging. We note that theintricately folded wings of a ladybird beetle can be deployed rapidly andeffectively sustain aerodynamic forces during flight; these abilitiesoriginate from the geometry and deformation of a specialized vein in the wingof this insect. We report compliant origami inspired by the wing vein inladybird beetles. The deformation and geometry of the compliant facet enablesboth large energy storage and self-locking in a single origami joint. On thebasis of our compliant origami, we developed a deployable glider module for amultimodal robot. The glider module is compactly foldable, is rapidlydeployable, and can effectively sustain aerodynamic forces. We also apply ourcompliant origami to enhance the energy storage capacity of the jumpingmechanism in a jumping robot.
| |
5_42.png | ONLINE COVER Space Hiking. NASA's Mars exploration rover, Spirit, traveled over 6.5 kilometers on Mars before its right-front drive wheel failed. Spirit's wheel would remain locked in low cohesion sulfate sands; this lack of mobility signaled the end of Spirit's mission. In 2015 NASA developed the Resource Prospector 15 (RP15), a planetary rover with a drive-train that combines wheel spinning with cyclic legged gaits. Shrivastava et al. built a laboratory model of RP15, called Mini Rover, to test the locomotion principles of RP15 and demonstrate hill climbing on loose granular media. This month's cover is a photograph of the Mini Rover. [CREDIT: CHRISTOPHER MOORE/INSTITUTE COMMUNICATIONS, GEORGIA INSTITUTE OF TECHNOLOGY, GEORGIA TECH RESEARCH CORPORATION] | science | Autonomous robots and vehicles must occasionally recover from locomotionfailure in loosely consolidated granular terrain. Recent mobility challengesled NASA Johnson Space Center to develop a prototype robotic lunar roverResource Prospector 15 (RP15) capable of wheeled, legged, and crawlingbehavior. To systematically understand the terradynamic performance of such adevice, we developed a scaled-down rover robot and studied its locomotion onslopes of dry and wet granular media. Addition of a cyclic-legged gait to therobots wheel spinning action changes the robot dynamics from that of awheeled vehicle to a locomotor paddling through frictional fluid. Granulardrag force measurements and modified resistive force theory facilitatemodeling of such dynamics. A peculiar gait strategy that agitates andcyclically reflows grains under the robot allows it to swim up looselyconsolidated hills. Whereas substrate disturbance typically hinders locomotionin granular media, the multimode design of RP15 and a diversity of possiblegaits facilitate formation of self-organized localized frictional fluids thatenable effective robust transport.
| |
5_45.png | ONLINE COVER RoBeetle Flexes Its Muscle. Untethered, small-scale robots are often powered by miniature batteries with low specific energy (compared with animal fat and chemical fuels). Inspired by the metabolism of animals, Yang et al. developed a robot, called RoBeetle, with size and mass comparable to those of a small insect that achieves crawling locomotion using an artificial micromuscle powered by the controlled catalytic combustion of methanol. RoBeetle can carry payloads up to 2.6 times its body weight, crawl on rough surfaces, and clamber up inclines of 15. This month's cover is a photograph of RoBeetle on a leaf (see also the Focus by Truby et al.). [CREDIT: XIUFENG YANG AND NSTOR O. PREZ-ARANCIBIA] | science | The creation of autonomous subgram microrobots capable of complex behaviorsremains a grand challenge in robotics largely due to the lack ofmicroactuators with high work densities and capable of using power sourceswith specific energies comparable to that of animal fat (38 megajoules perkilogram). Presently, the vast majority of microrobots are driven byelectrically powered actuators; consequently, because of the low specificenergies of batteries at small scales (below 1.8 megajoules per kilogram),almost all the subgram mobile robots capable of sustained operation remaintethered to external power sources through cables or electromagnetic fields.Here, we present RoBeetle, an 88-milligram insect-sized autonomous crawlingrobot powered by the catalytic combustion of methanol, a fuel with highspecific energy (20 megajoules per kilogram). The design and physicalrealization of RoBeetle is the result of combining the notion of controllableNiTi-Ptbased catalytic artificial micromuscle with that of integratedmillimeter-scale mechanical control mechanism (MCM). Through tetheredexperiments on several robotic prototypes and system characterization of thethermomechanical properties of their driving artificial muscles, we obtainedthe design parameters for the MCM that enabled RoBeetle to achieve autonomouscrawling. To evaluate the functionality and performance of the robot, weconducted a series of locomotion tests: crawling under two differentatmospheric conditions and on surfaces with different levels of roughness,climbing of inclines with different slopes, transportation of payloads, andoutdoor locomotion.
Continued development of untethered insect-scale robots will requirecodesigned power and actuation strategies.
| |
5_46.png | ONLINE COVER The Hannes Hand. Developing prostheses with improved functionality and better embodiment may reduce the abandonment rate of these devices. Laffranchi et al. report a bioinspired prosthetic upper limb system, called Hannes, that exhibits key properties of the human hand: accurate anthropomorphism, biomimetic performance, and human-like grasping behavior. Pilot clinical trials involving amputees showed that Hannes performed better than existing devices. This month's cover is a photograph of the Hannes hand prosthesis (see also the Focus by Nazarpour). [IMAGE CREDIT: DDPSTUDIO ARCHITECTURE & DESIGN] | science | Replacing the human hand with artificial devices of equal capability andeffectiveness is a long-standing challenge. Even the most advanced handprostheses, which have several active degrees of freedom controlled by theelectrical signals of the stumps residual muscles, do not achieve thecomplexity, dexterity, and adaptability of the human hand. Thus, prosthesisabandonment rate remains high due to poor embodiment. Here, we report aprosthetic hand called Hannes that incorporates key biomimetic properties thatmake this prosthesis uniquely similar to a human hand. By means of an holisticdesign approach and through extensive codevelopment work involvingresearchers, patients, orthopaedists, and industrial designers, our proposeddevice simultaneously achieves accurate anthropomorphism, biomimeticperformance, and human-like grasping behavior that outperform what is requiredin the execution of activities of daily living (ADLs). To evaluate theeffectiveness and usability of Hannes, pilot trials on amputees wereperformed. Tests and questionnaires were used before and after a period ofabout 2 weeks, in which amputees could autonomously use Hannes domestically toperform ADLs. Last, experiments were conducted to validate Hanness highperformance and the human likeness of its grasping behavior. Although Hannessspeed is still lower than that achieved by the human hand, our experimentsshowed improved performance compared with existing research or commercialdevices.
Co-creation leads the way for bioinspired prosthetics with improved design andperformance.
| |
5_47.png | ONLINE COVER A Trot in the Dark. The ability to traverse deformable surfaces (mud or snow) and dynamic footholds (rubble), as well as handle impediments such as thick vegetation and flowing water, is key to successfully navigating unstructured natural environments. Lee et al. have developed a locomotion controller that uses deep reinforcement learning to teach a quadruped robot how to navigate unseen and unstructured environments without the need for external sensors. The quadruped, ANYmal, was deployed in various outdoor settings to demonstrate that it could robustly traverse a range of challenging terrain relying solely on proprioception. This month's cover is a photograph of ANYmal atop Tenner Chrz, Tenna, Switzerland (see also the Focus by Ha). [IMAGE CREDIT: SIMON TANNER/NZZ] | science | Legged locomotion can extend the operational domain of robots to some of themost challenging environments on Earth. However, conventional controllers forlegged locomotion are based on elaborate state machines that explicitlytrigger the execution of motion primitives and reflexes. These designs haveincreased in complexity but fallen short of the generality and robustness ofanimal locomotion. Here, we present a robust controller for blind quadrupedallocomotion in challenging natural environments. Our approach incorporatesproprioceptive feedback in locomotion control and demonstrates zero-shotgeneralization from simulation to natural environments. The controller istrained by reinforcement learning in simulation. The controller is driven by aneural network policy that acts on a stream of proprioceptive signals. Thecontroller retains its robustness under conditions that were never encounteredduring training: deformable terrains such as mud and snow, dynamic footholdssuch as rubble, and overground impediments such as thick vegetation andgushing water. The presented work indicates that robust locomotion in naturalenvironments can be achieved by training in simple domains.
Deep reinforcement learning enables quadruped robots to traverse challengingnatural environments using only proprioception.
| |
5_48.png | ONLINE COVER Smooth Manipulator. Robot picking for commercial applications requires smooth robot arm motions from a variety of configurations to be computed quickly and efficiently. Ichnowski et al. have developed an algorithm for robot picking that considers jerk limits to produce smoother trajectories and uses deep learning to speed up the computation of those trajectories. The algorithm, which was validated on a UR5 robot, reduced the computation time of a preexisting motion planner from 25 seconds to 80 milliseconds. This month's cover is a photograph of a UR5 robot picking objects. [IMAGE CREDIT: ADAM LAU/BERKELEY ENGINEERING] | science | Robots for picking in e-commerce warehouses require rapid computing ofefficient and smooth robot arm motions between varying configurations. Recentresults integrate grasp analysis with arm motion planning to compute optimalsmooth arm motions; however, computation times on the order of tens of secondsdominate motion times. Recent advances in deep learning allow neural networksto quickly compute these motions; however, they lack the precision required toproduce kinematically and dynamically feasible motions. While infeasible, thenetwork-computed motions approximate the optimized results. The proposedmethod warm starts the optimization process by using the approximate motionsas a starting point from which the optimizing motion planner refines to anoptimized and feasible motion with few iterations. In experiments, theproposed deep learningbased warm-started optimizing motion planner reducescompute and motion time when compared to a sampling-based asymptoticallyoptimal motion planner and an optimizing motion planner. When applied tograsp-optimized motion planning, the results suggest that deep learning canreduce the computation time by two orders of magnitude (300), from 29 s to 80ms, making it practical for e-commerce warehouse picking.
| |
5_49.png | ONLINE COVER You Can Teach a Robot Dog New Tricks. Learning to adapt in unknown situations is key for robots to operate effectively in the wild. Inspired by the biomechanical control of muscular systems, Yang et al. developed a framework based on machine learning called multi-expert learning architecture to teach multiskill locomotion to a quadruped robot. Their controller consists of eight deep neural networks that represent expert skills that are then combined using a gated neural network to achieve complex locomotion such as coherent trotting, steering, and fall recovery. This month's cover is a photograph of Jueying, a quadruped robot, demonstrating agile motion. [IMAGE CREDIT: KAI YUAN AND CHRISTOPHER MCGREAVY] | science | Achieving versatile robot locomotion requires motor skills that can adapt topreviously unseen situations. We propose a multi-expert learning architecture(MELA) that learns to generate adaptive skills from a group of representativeexpert skills. During training, MELA is first initialized by a distinct set ofpretrained experts, each in a separate deep neural network (DNN). Then, bylearning the combination of these DNNs using a gating neural network (GNN),MELA can acquire more specialized experts and transitional skills acrossvarious locomotion modes. During runtime, MELA constantly blends multiple DNNsand dynamically synthesizes a new DNN to produce adaptive behaviors inresponse to changing situations. This approach leverages the advantages oftrained expert skills and the fast online synthesis of adaptive policies togenerate responsive motor skills during the changing tasks. Using one unifiedMELA framework, we demonstrated successful multiskill locomotion on a realquadruped robot that performed coherent trotting, steering, and fall recoveryautonomously and showed the merit of multi-expert learning generatingbehaviors that can adapt to unseen scenarios.
| |
6_50.png | ONLINE COVER Go Fish! Colonies of insects, flocks of birds, and schools of fish are able to produce complex global behaviors through simple local interactions. Inspired by nature, Berlinger et al. achieve complex 3D collective behavior with fish-inspired robots called Bluebots. Using implicit vision-based coordination, a swarming school of Bluebots, called Blueswarm, demonstrated synchrony, aggregation-dispersion, dynamic circle formation, and search-capture behaviors. This month's cover is a composite photograph of a Blueswarm (see also the Focus by Wolek et al.) [IMAGE CREDIT: FLORIAN BERLINGER] | science | Many fish species gather by the thousands and swim in harmony with seeminglyno effort. Large schools display a range of impressive collective behaviors,from simple shoaling to collective migration and from basic predator evasionto dynamic maneuvers such as bait balls and flash expansion. A wealth ofexperimental and theoretical work has shown that these complex three-dimensional (3D) behaviors can arise from visual observations of nearbyneighbors, without explicit communication. By contrast, most underwater robotcollectives rely on centralized, above-water, explicit communication and, as aresult, exhibit limited coordination complexity. Here, we demonstrate 3Dcollective behaviors with a swarm of fish-inspired miniature underwater robotsthat use only implicit communication mediated through the production andsensing of blue light. We show that complex and dynamic 3D collectivebehaviorssynchrony, dispersion/aggregation, dynamic circle formation, andsearch-capturecan be achieved by sensing minimal, noisy impressions ofneighbors, without any centralized intervention. Our results provide insightsinto the power of implicit coordination and are of interest for futureunderwater robots that display collective capabilities on par with fishschools for applications such as environmental monitoring and search in coralreefs and coastal environments.
A swarm of agile fish-robots uses vision-based implicit coordination todemonstrate self-organizing behaviors in a laboratory tank.
| |
6_51.png | ONLINE COVER Untethered Soft Robots Walk This Way. Pneumatic soft robots typically require bulky electromechanical components, such as pumps and valves, to achieve legged locomotion. To simplify the design of untethered quadruped robots, Drotman et al. use fluidic circuits, rather than electronic circuits, to achieve programmable locomotion in their robot. Specifically, a bioinspired gait pattern is achieved using a soft ring oscillator that produces rhythmic motions analogous to biological central pattern generator neural circuits found in nature. This month's cover is a photograph of a Drotman et al. legged soft robot (see also the Focus by Rajappan et al.) [CREDIT: DROTMAN ET AL./SCIENCE ROBOTICS] | science | Pneumatically actuated soft robots have recently shown promise for theirability to adapt to their environment. Previously, these robots have beencontrolled with electromechanical components, such as valves and pumps, thatare typically bulky and expensive. Here, we present an approach forcontrolling the gaits of soft-legged robots using simple pneumatic circuitswithout any electronic components. This approach produces locomotive gaitsusing ring oscillators composed of soft valves that generate oscillatingsignals analogous to biological central pattern generator neural circuits,which are acted upon by pneumatic logic components in response to sensorinputs. Our robot requires only a constant source of pressurized air to powerboth control and actuation systems. We demonstrate this approach by designingpneumatic control circuits to generate walking gaits for a soft-leggedquadruped with three degrees of freedom per leg and to switch between gaits tocontrol the direction of locomotion. In experiments, we controlled a basicwalking gait using only three pneumatic memory elements (valves). With twooscillator circuits (seven valves), we were able to improve locomotion speedby 270%. Furthermore, with a pneumatic memory element we designed to mimic adouble-pole double-throw switch, we demonstrated a control circuit thatallowed the robot to select between gaits for omnidirectional locomotion andto respond to sensor input. This work represents a step toward fullyautonomous, electronics-free walking robots for applications including low-cost robotics for entertainment and systems for operation in environmentswhere electronics may not be suitable.
A four-legged soft robot walks, rotates, and reacts to environmental obstaclesby incorporating a soft pneumatic control circuit.
| |
6_52.png | ONLINE COVER Microrobotic Shapeshifters. Robust, fast, and low-power actuation at the microscale is necessary for the continued advancement of microrobotic systems. Liu et al. have developed a micrometer-sized shape-memory actuator that achieves reversible actuation and shape-memory effects through surface electrochemical reactions. These programmable shape-memory actuators have a radius of curvature of 500 nanometers and an operation speed of less than 100 milliseconds and can be driven at voltages as low as 1 volt. This month's cover is a false-color scanning electron microscope image of an origami duck composed of microscale surface electrochemical actuators (see also the Focus by Omar et al.). [CREDIT: QINGKUN LIU/MCEUEN AND COHEN LABS/CORNELL UNIVERSITY] | science | Shape-memory actuators allow machines ranging from robots to medical implantsto hold their form without continuous power, a feature especially advantageousfor situations where these devices are untethered and power is limited.Although previous work has demonstrated shape-memory actuators using polymers,alloys, and ceramics, the need for micrometer-scale electroshape-memoryactuators remains largely unmet, especially ones that can be driven bystandard electronics (1 volt). Here, we report on a new class of fast, high-curvature, low-voltage, reconfigurable, micrometer-scale shape-memoryactuators. They function by the electrochemical oxidation/reduction of aplatinum surface, creating a strain in the oxidized layer that causes bending.They bend to the smallest radius of curvature of any electrically controlledmicroactuator (500 nanometers), are fast (<100-millisecond operation), andoperate inside the electrochemical window of water, avoiding bubble generationassociated with oxygen evolution. We demonstrate that these shape-memoryactuators can be used to create basic electrically reconfigurable microscalerobot elements including actuating surfaces, origami-based three-dimensionalshapes, morphing metamaterials, and mechanical memory elements. Our shape-memory actuators have the potential to enable the realization of adaptivemicroscale structures, bio-implantable devices, and microscopic robots.
Microscale programmable shape-memory actuators based on reversibleelectrochemical reactions can provide exciting opportunities formicrorobotics.
| |
6_53.png | ONLINE COVER A Wheel Advance. Transformable wheels based on membrane origami have the potential for large shape variation, high weight-to-payload ratio, and simple fabrication. However, existing membrane origami wheels have limited a load-bearing capacity. By introducing a wireframe design rule that accomodates thick membranes, Lee et al. have developed a high payload membrane origami wheel capable of bearing more than a 10-kN load. The load-bearing property of the wheel design was validated in a passenger vehicle field test. This month's cover is an illustration of a high load capacity origami transformable wheel. Credit: SOFT ROBOTICS RESEARCH CENTER, SNU | science | Composite membrane origami has been an efficient and effective method forconstructing transformable mechanisms while considerably simplifying theirdesign, fabrication, and assembly; however, its limited load-bearingcapability has restricted its application potential. With respect to wheeldesign, membrane origami offers unique benefits compared with its conventionalcounterparts, such as simple fabrication, high weight-to-payload ratio, andlarge shape variation, enabling softness and flexibility in a kinematicmechanism that neutralizes joint distortion and absorbs shocks from theground. Here, we report a transformable wheel based on membrane origamicapable of bearing more than a 10-kilonewton load. To achieve a high payload,we adopt a thick membrane as an essential element and introduce a wireframedesign rule for thick membrane accommodation. An increase in the thickness cancause a geometric conflict for the facet and the membrane, but the excessivestrain energy accumulation is unique to the thickness increase of themembrane. Thus, the design rules for accommodating membrane thickness aim toaddress both geometric and physical characteristics, and these rules areapplied to basic origami patterns to obtain the desired wheel shapes andtransformation. The capability of the resulting wheel applied to a passengervehicle and validated through a field test. Our study shows that membraneorigami can be used for high-payload applications.
| |
6_54.png | ONLINE COVER Robot Grasping and Manipulation. Considerable progress in robotic grasping has been made in picking and sorting regular objects in controlled environments. Yet, compared with human hands, which are exceptionally skilled at handling and transforming objects of various shapes and sizes, robotic hands remain inferior. For example, grasping and manipulation of soft, free-form objects remains difficult for robots. In this special issue, we seek to achieve better robotic grasping by considering the complex interplay of effective manipulator design, vision, perception (e.g., tactile), multimodal scene understanding, spatial planning, as well as learning and reasoning. This month's cover is a photograph of an unconventional two-fingered dexterous hand, Model W, developed by Bircher et al. that utilizes all surfaces of its fingers and palm for manipulation. Credit: BIRCHER ET AL./SCIENCE ROBOTICS | science | Humans use all surfaces of the hand for contact-rich manipulation. Robothands, in contrast, typically use only the fingertips, which can limitdexterity. In this work, we leveraged a potential energybased whole-handmanipulation model, which does not depend on contact wrench modeling liketraditional approaches, to design a robotic manipulator. Inspired by roboticcaging grasps and the high levels of dexterity observed in human manipulation,a metric was developed and used in conjunction with the manipulation model todesign a two-fingered dexterous hand, the Model W. This was accomplished bysimulating all planar finger topologies composed of open kinematic chains ofup to three serial revolute and prismatic joints, forming symmetric two-fingered hands, and evaluating their performance according to the metric. Wepresent the best design, an unconventional robot hand capable of performingcontinuous object reorientation, as well as repeatedly alternating betweenpower and pinch graspstwo contact-rich skills that have often eluded robotichandsand we experimentally characterize the hands manipulation capability.This hand realizes manipulation motions reminiscent of thumbindex fingermanipulative movement in humans, and its topology provides the foundation fora general-purpose dexterous robot hand.
| |
6_55.png | ONLINE COVER Groundbreaking Soft Robot. Robotic burrowing presents different challenges to robotic walking, swimming, or flying. Burrowing through soil requires subterranean interaction forces to be overcome; these forces can be an order of magnitude higher than air or water. Inspired by burrowing animals such as the southern sand octopus and the sandfish lizard, Naclerio et al. have developed a soft robot that can burrow through shallow dry sand with high speed and maneuverability by leveraging tip extension, granular fluidization, and tip asymmetries. Their robot can travel as fast as 480 cm/s and steer in three dimensions in shallow dry sand. This month's cover is a photograph of the burrowing soft robot emerging from the sand (see also the Focus by Tao). [CREDIT: SICHENG WANG] | science | Robotic navigation on land, through air, and in water is well researched;numerous robots have successfully demonstrated motion in these environments.However, one frontier for robotic locomotion remains largely unexploredbelowground. Subterranean navigation is simply hard to do, in part because theinteraction forces of underground motion are higher than in air or water byorders of magnitude and because we lack for these interactions a robustfundamental physics understanding. We present and test three hypotheses,derived from biological observation and the physics of granular intrusion, anduse the results to inform the design of our burrowing robot. These resultsreveal that (i) tip extension reduces total drag by an amount equal to theskin drag of the body, (ii) granular aeration via tip-based airflow reducesdrag with a nonlinear dependence on depth and flow angle, and (iii) variationof the angle of the tip-based flow has a nonmonotonic effect on lift ingranular media. Informed by these results, we realize a steerable, root-likesoft robot that controls subterranean lift and drag forces to burrow fasterthan previous approaches by over an order of magnitude and does so throughreal sand. We also demonstrate that the robot can modulate its pullout forceby an order of magnitude and control its direction of motion in both thehorizontal and vertical planes to navigate around subterranean obstacles. Ourresults advance the understanding and capabilities of robotic subterraneanlocomotion.
A bioinspired soft robot burrows through shallow dry sand with remarkablespeed and maneuverability.
| |
6_57.png | ONLINE COVER Swimming with AgnathaX. Experimentally investigating how local pressure-sensitive receptors and central pattern generators influence motor commands in vertebrate swimmers, such as lamprey, is difficult. To study undulatory swimming in these animals, Thandiackal et al. built a lamprey-like robot that captures the key components of a lamprey's neuromechanical system. Studies into their robot, called AgnathaX, combined with simulations showed that hydrodynamic force feedback contributes to motor pattern modulation as well as to motor rhythm generation and coordination. This month's cover is a photograph of a AgnathaX swimming in a body of water (see also Focus by Tytell and Long). [CREDIT: KAMILO MELO/BIOROBOTICS LABORATORYEPFL, KM-ROBOTA SRL, AND JAIR TORRES COY/KM- ROBOTA S.A.S] | science | Undulatory swimming represents an ideal behavior to investigate locomotioncontrol and the role of the underlying central and peripheral components inthe spinal cord. Many vertebrate swimmers have central pattern generators andlocal pressure-sensitive receptors that provide information about thesurrounding fluid. However, it remains difficult to study experimentally howthese sensors influence motor commands in these animals. Here, using aspecifically designed robot that captures the essential components of theanimal neuromechanical system and using simulations, we tested the hypothesisthat sensed hydrodynamic pressure forces can entrain body actuation throughlocal feedback loops. We found evidence that this peripheral mechanism leadsto self-organized undulatory swimming by providing intersegmental coordinationand body oscillations. Swimming can be redundantly induced by centralmechanisms, and we show that, therefore, a combination of both central andperipheral mechanisms offers a higher robustness against neural disruptionsthan any of them alone, which potentially explains how some vertebrates retainlocomotor capabilities after spinal cord lesions. These results broaden ourunderstanding of animal locomotion and expand our knowledge for the design ofrobust and modular robots that physically interact with the environment.
Skin sensors on an eel-like robot couple external hydrodynamic pressure withinternal neural patterns for robust swimming.
| |
6_58.png | ONLINE COVER Special Issue on Human-Robot Interaction. Prosthetic systems that provide a strong sense of agency and ownership between the user and the prosthesis will improve performance and reduce the rate of device abandonment. Marasco et al. have developed a neural-machine interface for a bionic prosthetic arm that leverages the sense of touch to enhance physical ownership and provides grip kinesthesia to enhance agency. This month's cover is a photograph of the inside socket of a bionic prosthetic arm (see also Focus by Ortiz-Catalan). Credit: Courtney Shell, Dylan Beckler, Zachary Thumser, and Paul Marasco/Laboratory for Bionic Integration, Cleveland Clinic | science | Bionic prostheses have restorative potential. However, the complex interplaybetween intuitive motor control, proprioception, and touch that represents thehallmark of human upper limb function has not been revealed. Here, we showthat the neurorobotic fusion of touch, grip kinesthesia, and intuitive motorcontrol promotes levels of behavioral performance that are stratified towardable-bodied function and away from standard-of-care prosthetic users. This wasachieved through targeted motor and sensory reinnervation, a closed-loopneural-machine interface, coupled to a noninvasive robotic architecture.Adding touch to motor control improves the ability to reach intended targetgrasp forces, find target durometers among distractors, and promote prostheticownership. Touch, kinesthesia, and motor control restore balanced decisionstrategies when identifying target durometers and intrinsic visuomotorbehaviors that reduce the need to watch the prosthetic hand during objectinteractions, which frees the eyes to look ahead to the next planned action.The combination of these three modalities also enhances error correctionperformance. We applied our unified theoretical, functional, and clinicalanalyses, enabling us to define the relative contributions of the sensory andmotor modalities operating simultaneously in this neural-machine interface.This multiperspective framework provides the necessary evidence to show thatbionic prostheses attain more human-like function with effective sensory-motorrestoration.
Integrating tactile and kinesthetic feedback in a bionic arm results inperformance closer to able-bodied individuals.
| |
6_59.png | ONLINE COVER A Hop, Skip, and a Jump. Birds and insects can switch between flying and walking with ease. Multimodal locomotion allows these animals to choose the most sensible mode of travel based on their environment. For robots, demonstrating either legged or aerial locomotion is challenging, let alone achieving both. Kim et al. have developed a bipedal robot that seamlessly combines legged and aerial locomotion. Their robot, called LEONARDO, is composed of light-weight multijoint legs and propellers. Synchronized control of LEONARDOs legs and propellers enable maneuvers that require delicate balancing, such as slackline walking and skateboarding. This month's cover is a photograph of LEONARDO balancing on one leg. See also the Focus by Mintchev. Credit: Elena-Sorina Lupu, Patrick Spieler, Kyunam Kim, and Soon-Jo Chung | science | Numerous mobile robots in various forms specialize in either ground or aeriallocomotion, whereas very few robots can perform complex locomotion tasksbeyond simple walking and flying. We present the design and control of amultimodal locomotion robotic platform called LEONARDO, which bridges the gapbetween two different locomotion regimes of flying and walking usingsynchronized control of distributed electric thrusters and a pair ofmultijoint legs. By combining two distinct locomotion mechanisms, LEONARDOachieves complex maneuvers that require delicate balancing, such as walking ona slackline and skateboarding, which are challenging for existing bipedalrobots. LEONARDO also demonstrates agile walking motions, interlaced withflying maneuvers to overcome obstacles using synchronized control ofpropellers and leg joints. The mechanical design and synchronized controlstrategy achieve a unique multimodal locomotion capability that couldpotentially enable robotic missions and operations that would be difficult forsingle-modal locomotion robots.
A multimodal robot demonstrates mobility and balance with the synchronous useof legs and propellers.
| |
6_60.png | ONLINE COVER Deep Dive. Long-term monitoring of deep ocean processes will lead to a better understanding of the oceanic carbon cycle. Smith et al. have developed a deep sea rover capable of measuring water temperature, oxygen concentration, current velocity and sediment community oxygen consumption at abyssal depths. The deep sea mobile robot, named Benthic Rover II, was deployed in the northeast Pacific Ocean in 1-year intervals for over 5 years at a depth of 4000 m. This month's cover is an illustration of the Benthic Rover II in action. Credit: Frame 48 Copyright 2020 MBARI | science | The deep-ocean carbon cycle is poorly quantified. An abyssal benthic rover wasdeveloped to make long time-series measurements of seafloor processes relatedto organic carbon remineralization and sequestration. Benthic Rover II (BR-II)is an autonomous dual-tracked vehicle that measures bottom water temperatureand oxygen concentration, current velocity, and sediment community oxygenconsumption (SCOC; respiration). BR-II is programmed to transit with lowsurface-contact pressure across the seafloor, photograph bottom conditions,and stop regularly to occupy respirometer incubation sites, with deploymentperiods up to 1 year. Now, continuously operational at a 4000-m station in thenortheast Pacific over 5 years, substantial weekly, seasonal, annual, andepisodic events have been recorded, which are critical to assessing the deep-ocean carbon cycle. There was a significant increase in phytodetritus cover (P < 0.01) arriving on the seafloor from the overlying water column between2015 and 2020 that was negatively correlated with bottom water dissolvedoxygen concentration ( P < 0.01). Over the continuous 5-year monitoringperiod from November 2015 to November 2020, SCOC was positively correlatedwith phytodetritus cover ( P < 0.01) and increased significantly from 2015to 2020 ( P < 0.01). These results show important influences of biologicalprocesses on the carbon cycle. The demonstrated success of BR-II now createsopportunities to expand the long-term monitoring of the deep sea to resolvethe coupling of water column and benthic processes key to understanding theoceanic carbon cycle on a planet engulfed in a changing climate.
| |
6_61.png | ONLINE COVER Takeoff and Landings. The ability for aerial robots to take off and land on a variety of irregular surfaces will broaden the range of environments where these robots could be deployed. Influenced by how birds take off, land, and grasp, Roderick et al. have developed a bird-inspired grasping system called Stereotyped Nature-inspired Aerial Grasper (SNAG). The researchers integrated SNAG into a quadcopter to demonstrate perching on tree branches of different sizes, orientations, and surface conditions. This month's cover is a photograph of SNAG on a quadcopter perched on a branch. Credit: William Roderick | science | Birds take off and land on a wide range of complex surfaces. In contrast,current robots are limited in their ability to dynamically grasp irregularobjects. Leveraging recent findings on how birds take off, land, and grasp, wedeveloped a biomimetic robot that can dynamically perch on complex surfacesand grasp irregular objects. To accommodate high-speed collisions, the robotstwo legs passively transform impact energy into grasp force, while theunderactuated grasping mechanism wraps around irregularly shaped objects inless than 50 milliseconds. To determine the range of hardware design,kinematic, behavior, and perch parameters that are sufficient for perchingsuccess, we launched the robot at tree branches. The results corroborate ourmathematical model, which shows that larger isometrically scaled animals androbots must accommodate disproportionately larger angular momenta, relative totheir mass, to achieve similar landing performance. We find that closed-loopbalance control serves an important role in maximizing the range of parameterssufficient for perching. The performance of the robots biomimetic featuresattests to the functionality of their avian counterparts, and the robotenables us to study aspects of bird legs in ways that are infeasible in vivo.Our data show that pronounced differences in modern avian toe arrangements donot yield large changes in perching performance, suggesting that arborealperching does not represent a strong selection pressure among common bird toetopographies. These findings advance our understanding of the avian perchingapparatus and highlight design concepts that enable robots to perch on naturalsurfaces for environmental monitoring.
| |
7_62.png | ONLINE COVER Yutu-2 Makes Tracks. The lunar farside is an underexplored region of the Moon. On 3 January 2019, the Chang'E-4 lander, housing the Yutu-2 rover, touched down on the lunar farside in the Von Krmn crater inside the South Pole-Aitken basinthe deepest, largest, and oldest basin on the Moon. Ding et al. report the deployment of Yutu-2 rover on the lunar farside over a period of 2 years (25 lunar days); during this time the rover examined the lunar surfaces regolith, craters, and rocks. This months cover is a photograph of the lunar surface taken with Yutu-2s navigation camera. Credit: Beijing Aerospace Control Center (BACC) | science | The lunar nearside has been investigated by many uncrewed and crewed missions,but the farside of the Moon remains poorly known. Lunar farside exploration ischallenging because maneuvering rovers with efficient locomotion in harshextraterrestrial environment is necessary to explore geologicalcharacteristics of scientific interest. ChangE-4 mission successfullytargeted the Moons farside and deployed a teleoperated rover (Yutu-2) toexplore inside the Von Krmn crater, conveying rich information regardingregolith, craters, and rocks. Here, we report mobile exploration on the lunarfarside with Yutu-2 over the initial 2 years. During its journey, Yutu-2 hasexperienced varying degrees of mild slip and skid, indicating that the terrainis relatively flat at large scales but scattered with local gentle slopes.Cloddy soil sticking on its wheels implies a greater cohesion of the lunarsoil than encountered at other lunar landing sites. Further identificationresults indicate that the regolith resembles dry sand and sandy loam on Earthin bearing properties, demonstrating greater bearing strength than thatidentified during the Apollo missions. In sharp contrast to the sparsity ofrocks along the traverse route, small fresh craters with unilateral moldableejecta are abundant, and some of them contain high-reflectance materials atthe bottom, suggestive of secondary impact events. These findings hint atnotable differences in the surface geology between the lunar farside andnearside. Experience gained with Yutu-2 improves the understanding of thefarside of the Moon, which, in return, may lead to locomotion with improvedefficiency and larger range.
| |
3_24.png | ONLINE COVER A Soft Touch. Measuring and distinguishing in real time normal and shear forces are crucial for slip detection and interaction with fragile bodies. Boutry et al. developed a multilayer electronic skin that distinguished subtle differences in force direction with microstructures inspired by human skin and sunflower floret spirals. They covered the hand of a robotic arm with the e-skin and programmed a controller to stop the automatic downward movement of the arm when a certain amount of force was detected. The resulting sensory feedback allowed the system to respond quickly enough to avoid damaging lightweight objects. [CREDIT: A. FOUDEH, C. BOUTRY, M. NEGRE, M. JORDA, O. VARDOULIS, Z. BAO/STANFORD UNIVERSITY] | science | Tactile sensing is required for the dexterous manipulation of objects inrobotic applications. In particular, the ability to measure and distinguish inreal time normal and shear forces is crucial for slip detection andinteraction with fragile objects. Here, we report a biomimetic soft electronicskin (e-skin) that is composed of an array of capacitors and capable ofmeasuring and discriminating in real time both normal and tangential forces.It is enabled by a three-dimensional structure that mimics the interlockeddermis-epidermis interface in human skin. Moreover, pyramid microstructuresarranged along nature-inspired phyllotaxis spirals resulted in an e-skin withincreased sensitivity, minimal hysteresis, excellent cycling stability, andresponse time in the millisecond range. The e-skin provided sensing feedbackfor controlling a robot arm in various tasks, illustrating its potentialapplication in robotics with tactile feedback.
| |
7_64.png | ONLINE COVER Special Issue on Biomechanics and Exoskeleton Suits. The metabolic rate during walking can be minimized by providing assistance at the waist of individuals undergoing physical therapy. By focusing on the center of mass of an individual, the waist, Antonellis et al. develop a robotic tether system that can apply assistive forward forces with optimized magnitude and timing. The researchers show that providing assistance early in the step cycle can reduce the metabolic rate by almost half in healthy adults. This months cover is a photograph of an individual with the robotic tether attached to their waist while walking on a treadmill. Credit: Ryan Soderlin/University of Nebraska at Omaha | science | The metabolic rate of walking can be reduced by applying a constant forwardforce at the center of mass. It has been shown that the metabolically optimalconstant force magnitude minimizes propulsion ground reaction force at theexpense of increased braking. This led to the hypothesis that selectivelyassisting propulsion could lead to greater benefits. We used a robotic waisttether to evaluate the effects of forward forces with different timings andmagnitudes. Here, we show that it is possible to reduce the metabolic rate ofhealthy participants by 48% with a greater efficiency ratio of metabolic costreduction per unit of net aiding work compared with other assistive robots.This result was obtained using a sinusoidal force profile with peak timingduring the middle of the double support. The same timing could also reduce themetabolic rate in patients with peripheral artery disease. A model explainsthat the optimal force profile accelerates the center of mass into theinverted pendulum movement during single support. Contrary to the hypothesis,the optimal force timing did not entirely coincide with propulsion. Within thefield of wearable robotics, there is a trend to use devices to mimicbiological torque or force profiles. Such bioinspired actuation can haverelevant benefits; however, our results demonstrate that this is notnecessarily optimal for reducing metabolic rate.
| |
7_65.png | ONLINE COVER Special Issue on Robots in Society. Robots need to be able to successfully negotiate everyday obstacles, such as doors, in order to find widespread application in society. Ito et al. have developed an approach to teach a robot how to autonomously open and pass through a door. First, a teleoperator teaches the robot the mechanics of door opening. Then, using computer vision combined with deep predictive learning, the robot is able to autonomously negotiate a variety of doors (inward-opening, outward-opening, and open doors) using what it learned through the teleoperator. This month's cover is a photograph of a robot autonomously opening an outward-opening door. Credit: Hiroshi Ito/Hitachi, Ltd. | science | Robots need robust models to effectively perform tasks that humans do on adaily basis. These models often require substantial developmental costs tomaintain because they need to be adjusted and adapted over time. Deepreinforcement learning is a powerful approach for acquiring complex real-worldmodels because there is no need for a human to design the model manually.Furthermore, a robot can establish new motions and optimal trajectories thatmay not have been considered by a human. However, the cost of learning is anissue because it requires a huge amount of trial and error in the real world.Here, we report a method for realizing complicated tasks in the real worldwith low design and teaching costs based on the principle of prediction errorminimization. We devised a module integration method by introducing amechanism that switches modules based on the prediction error of multiplemodules. The robot generates appropriate motions according to the doorsposition, color, and pattern with a low teaching cost. We also show that bycalculating the prediction error of each module in real time, it is possibleto execute a sequence of tasks (opening door outward and passing through) bylinking multiple modules and responding to sudden changes in the situation andoperating procedures. The experimental results show that the method iseffective at enabling a robot to operate autonomously in the real world inresponse to changes in the environment.
| |
7_66.png | ONLINE COVER Special Issue on Robots in the Wild. Robots have been successfully deployed in a wide range of domainsincluding land, sea, air, and spacefor a variety of applications such as search and rescue, oceanography, wildlife surveys, and space exploration. In this issue, Zhou et al. have developed a trajectory planner for swarms of micro drones that can be implemented using only an onboard computer. Their planner computes trajectories based on limited information from the drone's onboard sensors to enable collision-free flight in cluttered environments in the wild. This month's cover is a photo illustration of a swarm of micro-drones flying through a forest (see also the Focus by Soria). Credit: Zhou et al./Zhejiang University | science | Aerial robots are widely deployed, but highly cluttered environments such asdense forests remain inaccessible to drones and even more so to swarms ofdrones. In these scenarios, previously unknown surroundings and narrowcorridors combined with requirements of swarm coordination can createchallenges. To enable swarm navigation in the wild, we develop miniature butfully autonomous drones with a trajectory planner that can function in atimely and accurate manner based on limited information from onboard sensors.The planning problem satisfies various task requirements including flightefficiency, obstacle avoidance, and inter-robot collision avoidance, dynamicalfeasibility, swarm coordination, and so on, thus realizing an extensibleplanner. Furthermore, the proposed planner deforms trajectory shapes andadjusts time allocation synchronously based on spatial-temporal jointoptimization. A high-quality trajectory thus can be obtained afterexhaustively exploiting the solution space within only a few milliseconds,even in the most constrained environment. The planner is finally integratedinto the developed palm-sized swarm platform with onboard perception,localization, and control. Benchmark comparisons validate the superiorperformance of the planner in trajectory quality and computing time. Variousreal-world field experiments demonstrate the extensibility of our system. Ourapproach evolves aerial robotics in three aspects: capability of clutteredenvironment navigation, extensibility to diverse task requirements, andcoordination as a swarm without external facilities.
| |
7_67.png | ONLINE COVER Special Issue on Electronic Skin and Computing Hardware for Robotics. In order for robots to be deployed effectively, they must be able to interact with their environment. Yu et al. have developed a printable electronic skin to endow robots with tactile and chemical sensing capabilities to detect physical objects and hazardous materials. The electronic skin can also be affixed to an operators arm to detect electromyography signals and to control the movement of a robot. This months cover is a photograph of the electronic skin placed on an operators arm and also incorporated in the fingers of a robot. Credit: Wei Gao/CalTech | science | Ultrasensitive multimodal physicochemical sensing for autonomous roboticdecision-making has numerous applications in agriculture, security,environmental protection, and public health. Previously reported roboticsensing technologies have primarily focused on monitoring physical parameterssuch as pressure and temperature. Integrating chemical sensors for autonomousdry-phase analyte detection on a robotic platform is rather extremelychallenging and substantially underdeveloped. Here, we introduce an artificialintelligencepowered multimodal robotic sensing system (M-Bot) with an all-printed mass-producible soft electronic skinbased human-machine interface. Ascalable inkjet printing technology with custom-developed nanomaterial inkswas used to manufacture flexible physicochemical sensor arrays forelectrophysiology recording, tactile perception, and robotic sensing of a widerange of hazardous materials including nitroaromatic explosives, pesticides,nerve agents, and infectious pathogens such as SARS-CoV-2. The M-Bot decodesthe surface electromyography signals collected from the human body throughmachine learning algorithms for remote robotic control and can perform in situthreat compound detection in extreme or contaminated environments with user-interactive tactile and threat alarm feedback. The printed electronicskinbased robotic sensing technology can be further generalized and appliedto other remote sensing platforms. Such diversity was validated on anintelligent multimodal robotic boat platform that can efficiently track thesource of trace amounts of hazardous compounds through autonomous andintelligent decision-making algorithms. This fully printed human-machineinteractive multimodal sensing technology could play a crucial role indesigning future intelligent robotic systems and can be easily reconfiguredtoward numerous practical wearable and robotic applications.
| |
7_68.png | ONLINE COVER A Model of Oneself. Robots, like humans and animals, require self-models to be able to anticipate and plan future actions. Chen et al. developed a method that enables a robot arm to model its morphology and kinematics using an approach based on query-driven visual self-modeling. This months cover is a multiple-exposure image of a self-modeled robot arm touching a small red sphere while avoiding a large red cube. Credit: Columbia Engineering | science | Internal computational models of physical bodies are fundamental to theability of robots and animals alike to plan and control their actions. Theseself-models allow robots to consider outcomes of multiple possible futureactions without trying them out in physical reality. Recent progress in fullydata-driven self-modeling has enabled machines to learn their own forwardkinematics directly from task-agnostic interaction data. However, forwardkinematic models can only predict limited aspects of the morphology, such asthe position of end effectors or velocity of joints and masses. A keychallenge is to model the entire morphology and kinematics without priorknowledge of what aspects of the morphology will be relevant to future tasks.Here, we propose that instead of directly modeling forward kinematics, a moreuseful form of self-modeling is one that could answer space occupancy queries,conditioned on the robots state. Such query-driven self-models are continuousin the spatial domain, memory efficient, fully differentiable, and kinematicaware and can be used across a broader range of tasks. In physicalexperiments, we demonstrate how a visual self-model is accurate to about 1% ofthe workspace, enabling the robot to perform various motion planning andcontrol tasks. Visual self-modeling can also allow the robot to detect,localize, and recover from real-world damage, leading to improved machineresiliency.
| |
7_69.png | ONLINE COVER Flexing Biomolecular Muscle. Actuation of soft microrobots can be achieved through the application of miniature motors, such as artificial muscles. However, the integration of such actuators within soft microrobots requires assembly processes that are often slow and require multiple steps. Wang et al. have developed a rapid and cost-effective in situ fabrication process that integrates artificial muscles composed of molecular motors in a variety of microrobots. This months cover shows a microrobot manipulating a ball; its arms are sequentially controlled by light-responsive artificial muscles. Credit: Wang et al./Osaka University | science | Microrobots have been developed for applications in the submillimeter domainsuch as the manipulation of micro-objects and microsurgery. Rapid progress hasbeen achieved in developing miniaturized components for microrobotic systems,resulting in a variety of functional microactuators and soft components forcreating untethered microrobots. Nevertheless, the integration ofmicrocomponents, especially the assembly of actuators and mechanicalcomponents, is still time-consuming and has inherent restrictions, thuslimiting efficient fabrications of microrobots and their potentialapplications. Here, we propose a method for fabricating microrobots in situinspired by the construction of microsystems in living organisms. In amicrofluidic chip, hydrogel mechanical components and artificial muscleactuators are successively photopatterned from hydrogel prepolymer andbiomolecular motors, respectively, and integrated in situ into functionalmicrorobots. The proposed method allows the fast fabrication of microrobotsthrough simple operations and affordable materials while providing versatilefunctions through the precise spatiotemporal control of in situ integrationand reconfiguration of artificial muscles. To validate the method, wefabricated microrobots to elicit different motions and on-chip robots withunique characteristics for microfluidic applications. This study may establisha new paradigm for microrobot integration and lead to the production of uniquebiohybrid microrobots with various advantages.
| |
7_70.png | ONLINE COVER Microrobots Make Big Steps Toward Autonomy. Existing microrobots are generally limited to external control through magnetic fields or light. Microrobots that are able to integrate control, power, and sensing internally (or onboard) are expected to achieve greater autonomy than their externally controlled counterparts. Reynolds et al. have built microscopic robots controlled by onboard complementary metal oxide semiconductor (CMOS) electronics. These microrobots are powered by light and use onboard computers to execute pre-programmed motions. This months cover is a false-color scanning electron microscope image of microscopic robots before release. Credit: Qingkun Liu/McEuen and Cohen Labs/Cornell University | science | Autonomous robotssystems where mechanical actuators are guided through aseries of states by information processing units to perform a predesignedfunctionare expected to revolutionize everything from health care totransportation. Microscopic robots are poised for a similar revolution infields from medicine to environmental remediation. A key hurdle to developingthese microscopic robots is the integration of information systems,particularly electronics fabricated at commercial foundries, withmicroactuators. Here, we develop such an integration process and buildmicroscopic robots controlled by onboard complementary metal oxidesemiconductor electronics. The resulting autonomous, untethered robots are 100to 250 micrometers in size, are powered by light, and walk at speeds greaterthan 10 micrometers per second. In addition, we demonstrate a microscopicrobot that can respond to an optical command. This work paves the way forubiquitous autonomous microscopic robots that perform complex functions,respond to their environments, and communicate with the outside world.
| |
7_71.png | ONLINE COVER Physical Connections. Morphological computing leverages the dynamics of mechanical bodies for computation with the aim to decentralize the computing load. Inspired by artificial neural networks, Lee et al. have realized a mechanical neural network that uses interconnected beams with tunable stiffness to learn mechanical behaviors. Genetic and partial pattern search algorithms were applied to the mechanical neural network. This month's cover is a photograph of the mechanical neural network in operation. Credit: Lee et al./Science Robotics | science | Aside from some living tissues, few materials can autonomously learn toexhibit desired behaviors as a consequence of prolonged exposure tounanticipated ambient loading scenarios. Still fewer materials can continue toexhibit previously learned behaviors in the midst of changing conditions(e.g., rising levels of internal damage, varying fixturing scenarios, andfluctuating external loads) while also acquiring new behaviors best suited forthe situation at hand. Here, we describe a class of architected materials,called mechanical neural networks (MNNs), that achieve such learningcapabilities by tuning the stiffness of their constituent beams similar to howartificial neural networks (ANNs) tune their weights. An example lattice wasfabricated to demonstrate its ability to learn multiple mechanical behaviorssimultaneously, and a study was conducted to determine the effect of latticesize, packing configuration, algorithm type, behavior number, and linear-versus-nonlinear stiffness tunability on MNN learning as proposed. Thus, thiswork lays the foundation for artificial-intelligent (AI) materials that canlearn behaviors and properties.
| |
7_72.png | ONLINE COVER Bionics replicate biomechanics. Lower-limb prostheses are often heavy, consume a lot of energy, and do not replicate the natural gait. Tran et al. have developed a powered prosthesis for above-knee amputees that possesses knee, ankle, and toe joints to replicate the biomechanics of the leg. The device is lightweight and capable of regenerating energy during walking to extend its battery life. The researchers validated their design with preclinical studies on above-knee amputees showing the ability to ambulate on level ground and on stairs. This months cover is a photograph of the prosthesis device: Utah Bionic Leg. Credit: Christoph Neumann & Sascha Boldt | Ottobock | science | Robotic leg prostheses promise to improve the mobility and quality of life ofmillions of individuals with lower-limb amputations by imitating thebiomechanics of the missing biological leg. Unfortunately, existing poweredprostheses are much heavier and bigger and have shorter battery life thanconventional passive prostheses, severely limiting their clinical viabilityand utility in the daily life of amputees. Here, we present a robotic legprosthesis that replicates the key biomechanical functions of the biologicalknee, ankle, and toe in the sagittal plane while matching the weight, size,and battery life of conventional microprocessor-controlled prostheses. Thepowered knee joint uses a unique torque-sensitive mechanism combining thebenefits of elastic actuators with that of variable transmissions. A singleactuator powers the ankle and toe joints through a compliant, underactuatedmechanism. Because the biological toe dissipates energy while the biologicalankle injects energy into the gait cycle, this underactuated systemregenerates substantial mechanical energy and replicates the key biomechanicalfunctions of the ankle/foot complex during walking. A compact prosthesis frameencloses all mechanical and electrical components for increased robustness andefficiency. Preclinical tests with three individuals with above-kneeamputation show that the proposed robotic leg prosthesis allows for commonambulation activities with close to normative kinematics and kinetics. Usingan optional passive mode, users can walk on level ground indefinitely withoutcharging the battery, which has not been shown with any other powered ormicroprocessor-controlled prostheses. A prosthesis with these characteristicshas the potential to improve real-world mobility in individuals with above-knee amputation.
| |
7_73.png | ONLINE COVER Robot on the Wall. Terrestrial robots that can operate on horizontal and vertical surfaces will enable more opportunities for their application in the real world. Hong et al. have developed a fast-climbing legged robot that can scale ferromagnetic walls and ceilings, maneuver over small obstacles, and carry payloads. The quadruped robot called MARVEL (Magnetically Adhesive Robot for Versatile and Expeditious Locomotion) climbs using specially designed feet based on electro-permanent magnets and magnetorheological elastomers. This month's cover is a photograph of MARVEL climbing on a ferromagnetic surface. Credit: Yong Um and Seungwoo Hong/Department of Mechanical Engineering, Korea Advanced Institute of Science and Technology | science | A climbing robot that can rapidly move on diverse surfaces such as floors,walls, and ceilings will have an enlarged operational workspace compared withother terrestrial robots. However, the climbing skill of robots in suchenvironments has been limited to low speeds or simple locomotion tasks. Here,we present an untethered quadrupedal climbing robot called MARVEL(magnetically adhesive robot for versatile and expeditious locomotion),capable of agile and versatile climbing locomotion in ferromagneticenvironments. MARVEL excels over prior climbing robots in terms of climbingspeed and ability to execute various motions. It demonstrates the fastestvertical and inverted walking speed, whereas its versatile locomotion abilityenables the highest number of gaits and locomotion tasks. The key innovationsare an integrated foot design using electropermanent magnets andmagnetorheological elastomers that provide large adhesion and traction forces,torque control actuators, and a model predictive control framework adapted forstable climbing. In experiments, the robot achieved locomotion on ceilings andvertical walls up to 0.5 meter (1.51 body lengths) per second and 0.7 meter(2.12 body lengths) per second, respectively. Furthermore, the robot exhibitedcomplex behaviors such as stepping over 10-centimeter-wide gaps; overcoming5-centimeter-high obstacles; and making transitions between floors, walls, andceilings. We also show that MARVEL could climb on a curved surface of astorage tank covered with up to 0.3-millimeter-thick paint with rust and dust.
| |
8_74.png | ONLINE COVER Drones branch out. Continuous and systematic biomonitoring of natural environments could enable a better understanding of how the biosphere responds to environmental changes. Environmental DNA (eDNA) is a useful tool for assessing biodiversity levels, particularly in forests. Aucone et al. report a drone that can autonomously land and sample eDNA on tree branches with different stiffnesses. This month's cover is a photograph of the drone resting on a branch. Credit: Gottardo Pestalozzi and Enrico Pestalozzi | science | The protection and restoration of the biosphere is crucial for humanresilience and well-being, but the scarcity of data on the status anddistribution of biodiversity puts these efforts at risk. DNA released into theenvironment by organisms, i.e., environmental DNA (eDNA), can be used tomonitor biodiversity in a scalable manner if equipped with the appropriatetool. However, the collection of eDNA in terrestrial environments remains achallenge because of the many potential surfaces and sources that need to besurveyed and their limited accessibility. Here, we propose to surveybiodiversity by sampling eDNA on the outer branches of tree canopies with anaerial robot. The drone combines a force-sensing cage with a haptic-basedcontrol strategy to establish and maintain contact with the upper surface ofthe branches. Surface eDNA is then collected using an adhesive surfaceintegrated in the cage of the drone. We show that the drone can autonomouslyland on a variety of branches with stiffnesses between 1 and 103 newton/meterwithout prior knowledge of their structural stiffness and with robustness tolinear and angular misalignments. Validation in the natural environmentdemonstrates that our method is successful in detecting animal species,including arthropods and vertebrates. Combining robotics with eDNA samplingfrom a variety of unreachable aboveground substrates can offer a solution forbroad-scale monitoring of biodiversity.
| |
8_75.png | ONLINE COVER Quick Reaction. Exoskeleton suits can improve standing balance by quickly reacting to postural perturbations. Beck et al. have developed a wearable ankle exoskeleton suit, ExoBoot, that can generate artificially fast torque before the onset of the users physiological reaction. ExoBoot was validated with human participants, preventing the users ankle joint from dorsiflexion and improving their standing balance when the surface they were stood on was displaced. This months cover is a multi-exposure image of a user wearing the ankle ExoBoot that is able to correct their standing balance following postural perturbation. Credit: Courtesy of Georgia Institute of Technology | science | Maintaining balance throughout daily activities is challenging because of theunstable nature of the human body. For instance, a persons delayed reactiontimes limit their ability to restore balance after disturbances. Wearableexoskeletons have the potential to enhance user balance after a disturbance byreacting faster than physiologically possible. However, artificially fastbalance-correcting exoskeleton torque may interfere with the users ensuingphysiological responses, consequently hindering the overall reactive balanceresponse. Here, we show that exoskeletons need to react faster thanphysiological responses to improve standing balance after posturalperturbations. Delivering ankle exoskeleton torque before the onset ofphysiological reactive joint moments improved standing balance by 9%, whereasdelaying torque onset to coincide with that of physiological reactive anklemoments did not. In addition, artificially fast exoskeleton torque disruptedthe ankle mechanics that generate initial local sensory feedback, but theinitial reactive soleus muscle activity was only reduced by 18% versusbaseline. More variance of the initial reactive soleus muscle activity wasaccounted for using delayed and scaled whole-body mechanics [specificallycenter of mass (CoM) velocity] versus local ankleor soleusfasciclemechanics, supporting the notion that reactive muscle activity iscommanded to achieve task-level goals, such as maintaining balance. Together,to elicit symbiotic human-exoskeleton balance control, device torque may needto be informed by mechanical estimates of global sensory feedback, such as CoMkinematics, that precede physiological responses.
| |
8_76.png | ONLINE COVER Robust robots. The ability of robots to endure severe damage, as well as offer the possibility for repair, can enable their deployment in challenging settings. Kim et al. have developed aerial robots that can endure several punctures while sustaining controlled flight. The dielectric elastomer actuators on the robots were also designed to be repairable by using laser ablation to isolate the defects and recover their performance. This months cover is a photograph of the aerial robot showing one dielectric elastomer actuator pierced with fiberglass needles. Credit: Yi-Husan Hsiao and Sampson Wilcox | science | Insects maintain remarkable agility after incurring severe injuries or wounds.Although robots driven by rigid actuators have demonstrated agile locomotionand manipulation, most of them lack animal-like robustness against unexpecteddamage. Dielectric elastomer actuators (DEAs) are a class of muscle-like softtransducers that have enabled nimble aerial, terrestrial, and aquatic roboticlocomotion comparable to that of rigid actuators. However, unlike muscles,DEAs suffer local dielectric breakdowns that often cause global devicefailure. These local defects severely limit DEA performance, lifetime, andsize scalability. We developed DEAs that can endure more than 100 punctureswhile maintaining high bandwidth (>400 hertz) and power density (>700 watt perkilogram)sufficient for supporting energetically expensive locomotion such asflight. We fabricated electroluminescent DEAs for visualizing electrodeconnectivity under actuator damage. When the DEA suffered severe dielectricbreakdowns that caused device failure, we demonstrated a laser-assisted repairmethod for isolating the critical defects and recovering performance. Theseresults culminate in an aerial robot that can endure critical actuator andwing damage while maintaining similar accuracy in hovering flight. Our workhighlights that soft robotic systems can embody animal-like agility andresiliencea critical biomimetic capability for future robots to interact withchallenging environments.
| |
8_77.png | ONLINE COVER Learning-based navigation. The ability of robots to navigate in unfamiliar environments presents a challenge for autonomous flight. Chahine et al. have developed and implemented a learning framework that utilizes liquid neural networks to facilitate the navigation of quadrotor robots to specific targets in surroundings that are distinct from their training environment. The brain-inspired framework relies on a learning-based solution that is robust and adaptable, in order to complete navigation tasks without guidance. This months cover is an image of a drone equipped with the liquid neural network performing vision-based autonomous flight toward a moving human target. Credit: Mike Grimmett/MIT Computer Science and Artificial Intelligence Laboratory/Copyright Massachusetts Institute of Technology 2023 | science | Autonomous robots can learn to perform visual navigation tasks from offlinehuman demonstrations and generalize well to online and unseen scenarios withinthe same environment they have been trained on. It is challenging for theseagents to take a step further and robustly generalize to new environments withdrastic scenery changes that they have never encountered. Here, we present amethod to create robust flight navigation agents that successfully performvision-based fly-to-target tasks beyond their training environment underdrastic distribution shifts. To this end, we designed an imitation learningframework using liquid neural networks, a brain-inspired class of continuous-time neural models that are causal and adapt to changing conditions. Weobserved that liquid agents learn to distill the task they are given fromvisual inputs and drop irrelevant features. Thus, their learned navigationskills transferred to new environments. When compared with several otherstate-of-the-art deep agents, experiments showed that this level of robustnessin decision-making is exclusive to liquid networks, both in their differentialequation and closed-form representations.
| |
8_78.png | ONLINE COVER Special issue on brain-inspired and brain-interfacing robots. The ability of robots to recognize places is crucial for interacting with their environment. Inspired by the multimodal sensory processing by the brain, Fangwen Yu et al. have developed a general place recognition system that relies on multiple sensors, in combination with multimodal hybrid neural network, and neuromorphic computing. The system was deployed on a legged robot and shown to be capable of accurately recognizing places indoors and outdoors. This months cover is an image of the place recognition system onboard a legged robot in a forest. Credit: Luping et al./Science Robotics | science | Place recognition is an essential spatial intelligence capability for robotsto understand and navigate the world. However, recognizing places in naturalenvironments remains a challenging task for robots because of resourcelimitations and changing environments. In contrast, humans and animals canrobustly and efficiently recognize hundreds of thousands of places indifferent conditions. Here, we report a brain-inspired general placerecognition system, dubbed NeuroGPR, that enables robots to recognize placesby mimicking the neural mechanism of multimodal sensing, encoding, andcomputing through a continuum of space and time. Our system consists of amultimodal hybrid neural network (MHNN) that encodes and integrates multimodalcues from both conventional and neuromorphic sensors. Specifically, to encodedifferent sensory cues, we built various neural networks of spatial viewcells, place cells, head direction cells, and time cells. To integrate thesecues, we designed a multiscale liquid state machine that can process and fusemultimodal information effectively and asynchronously using diverse neuronaldynamics and bioinspired inhibitory circuits. We deployed the MHNN on Tianjic,a hybrid neuromorphic chip, and integrated it into a quadruped robot. Ourresults show that NeuroGPR achieves better performance compared withconventional and existing biologically inspired approaches, exhibitingrobustness to diverse environmental uncertainty, including perceptualaliasing, motion blur, light, or weather changes. Running NeuroGPR as anoverall multineural network workload on Tianjic showcases its advantages with10.5 times lower latency and 43.6% lower power consumption than the commonlyused mobile robot processor Jetson Xavier NX.
| |
8_79.png | ONLINE COVER Soft Robot Fabrication. Fabricating fully functional soft robots that do not require assembly of the component parts can be achieved by monolithic three-dimensional printing. Zhai et al. have developed a 3D-printing process that can produce airtight soft robots integrated with pneumatic valves, control switches, and fluidic circuits that are fabricated within the grippers during the printing process. The electronics-free soft grippers were reliant on the integrated contact and gravity switches to control the capture and release of objects. This months cover is an image of a soft robot gripper, developed by the monolithic 3D-printing process, grasping a tennis ball. Credit: Zhai et al./Science Robotics | science | Most soft robots are pneumatically actuated and fabricated by molding andassembling processes that typically require many manual operations and limitcomplexity. Furthermore, complex control components (for example, electronicpumps and microcontrollers) must be added to achieve even simple functions.Desktop fused filament fabrication (FFF) three-dimensional printing providesan accessible alternative with less manual work and the capability ofgenerating more complex structures. However, because of material and processlimitations, FFF-printed soft robots often have a high effective stiffness andcontain a large number of leaks, limiting their applications. We present anapproach for the design and fabrication of soft, airtight pneumatic roboticdevices using FFF to simultaneously print actuators with embedded fluidiccontrol components. We demonstrated this approach by printing actuators anorder of magnitude softer than those previously fabricated using FFF andcapable of bending to form a complete circle. Similarly, we printed pneumaticvalves that control a high-pressure airflow with low control pressure.Combining the actuators and valves, we demonstrated a monolithically printedelectronics-free autonomous gripper. When connected to a constant supply ofair pressure, the gripper autonomously detected and gripped an object andreleased the object when it detected a force due to the weight of the objectacting perpendicular to the gripper. The entire fabrication process of thegripper required no posttreatment, postassembly, or repair of manufacturingdefects, making this approach highly repeatable and accessible. Our proposedapproach represents a step toward complex, customized robotic systems andcomponents created at distributed fabricating facilities.
| |
8_80.png | ONLINE COVER Special Issue on Autonomy and AI in Robotics. A team of legged robots can efficiently explore unstructured terrains with task-level autonomy. Arm et al. report on a robot team comprising a scout that can identify potential scientific targets in an environment, a hybrid that collects data from the targets, and a scientist that performs in-depth scientific analysis of the targets. The robot team could efficiently map terrain mimicking planetary environments, identify resource-enriched areas, and scientifically analyze targets of interest. This months cover is an image of a team of legged robots exploring a field of boulders. Credit: Arm et al./Science Robotics | science | The interest in exploring planetary bodies for scientific investigation and insitu resource utilization is ever-rising. Yet, many sites of interest areinaccessible to state-of-the-art planetary exploration robots because of therobots inability to traverse steep slopes, unstructured terrain, and loosesoil. In addition, current single-robot approaches only allow a limitedexploration speed and a single set of skills. Here, we present a team oflegged robots with complementary skills for exploration missions inchallenging planetary analog environments. We equipped the robots with anefficient locomotion controller, a mapping pipeline for online and postmissionvisualization, instance segmentation to highlight scientific targets, andscientific instruments for remote and in situ investigation. Furthermore, weintegrated a robotic arm on one of the robots to enable high-precisionmeasurements. Legged robots can swiftly navigate representative terrains, suchas granular slopes beyond 25, loose soil, and unstructured terrain,highlighting their advantages compared with wheeled rover systems. Wesuccessfully verified the approach in analog deployments at the Beyond GravityExoMars rover test bed, in a quarry in Switzerland, and at the Space ResourcesChallenge in Luxembourg. Our results show that a team of legged robots withadvanced locomotion, perception, and measurement skills, as well as task-levelautonomy, can conduct successful, effective missions in a short time. Ourapproach enables the scientific exploration of planetary target sites that arecurrently out of human and robotic reach.
| |
8_81.png | ONLINE COVER Estimating UncertaintyA Deep Learning Framework for Estimating Uncertainty in Optical Flow for Real-Time Control of Robots. Sanket, Singh et al. report on a generalized formulation, Ajna, that was deployed on an aerial robot with a single monocular camera and onboard computing. The approach was used for various robotic tasks, such as dodging dynamic obstacles, navigating static obstacles, and flying through unknown gaps, as well as computer vision tasks, such as segmentation of an unknown object pile. This months cover is a time-lapse image of a drone using the Ajna framework to navigate around static obstacles. Credit: Nitin J. Sanket/Perception and Autonomous Robotics Group, Worcester Polytechnic Institute | science | Robots are active agents that operate in dynamic scenarios with noisy sensors.Predictions based on these noisy sensor measurements often lead to errors andcan be unreliable. To this end, roboticists have used fusion methods usingmultiple observations. Lately, neural networks have dominated the accuracycharts for perception-driven predictions for robotic decision-making and oftenlack uncertainty metrics associated with the predictions. Here, we present amathematical formulation to obtain the heteroscedastic aleatoric uncertaintyof any arbitrary distribution without prior knowledge about the data. Theapproach has no prior assumptions about the prediction labels and is agnosticto network architecture. Furthermore, our class of networks, Ajna, addsminimal computation and requires only a small change to the loss functionwhile training neural networks to obtain uncertainty of predictions, enablingreal-time operation even on resource-constrained robots. In addition, we studythe informational cues present in the uncertainties of predicted values andtheir utility in the unification of common robotics problems. In particular,we present an approach to dodge dynamic obstacles, navigate through acluttered scene, fly through unknown gaps, and segment an object pile, withoutcomputing depth but rather using the uncertainties of optical flow obtainedfrom a monocular camera with onboard sensing and computation. We successfullyevaluate and demonstrate the proposed Ajna network on four aforementionedcommon robotics and computer vision tasks and show comparable results tomethods directly using depth. Our work demonstrates a generalized deepuncertainty method and demonstrates its utilization in robotics applications.
| |
7_63.png | ONLINE COVER Special Issue on Soft Systems and Sensors. Soft robots require a sustainable source of raw materials to facilitate recyclability and degradation following the end of their life cycle. Unlike the synthetic polymers widely used to date, Heiden et al. 3D-printed soft robots from gelatin and sugar, with the capacity to dissolve fully in water and degrade enzymatically. These soft robots can also be recycled and reprinted at least four more times using the same material without losing their mechanical stability. Furthermore, the integration of waveguides enables actuators to sense objects within their vicinity through changes in light intensity. This month's cover shows a melt extrusion-based printer nozzle depositing a gelatin-based hydrogel to fabricate a 3D structure of a soft robot (see also the Focus by Tan). Credit: A. Heiden et al./Johannes Kepler University Linz | science | Soft robotics greatly benefits from nature as a source of inspiration,introducing innate means of safe interaction between robotic appliances andliving organisms. In contrast, the materials involved are oftennonbiodegradable or stem from nonrenewable resources, contributing to an ever-growing environmental footprint. Furthermore, conventional manufacturingmethods, such as mold casting, are not suitable for replicating or imitatingthe complexity of natures creations. Consequently, the inclusion ofsustainability concepts alongside the development of new fabricationprocedures is required. We report a customized 3D-printing process based onfused deposition modeling, printing a fully biodegradable gelatin-basedhydrogel (biogel) ink into dimensionally stable, complex objects. This processenables fast and cost-effective prototyping of resilient, soft roboticapplications from gels that stretch to six times their original length, aswell as an accessible recycling procedure with zero waste. We present printedpneumatic actuators performing omnidirectional movement at fast response times(less than a second), featuring integrated 3D-printed stretchable waveguides,capable of both proprio- and exteroception. These soft devices are endowedwith dynamic real-time control capable of automated search-and-wipe routinesto detect and remove obstacles. They can be reprinted several times ordisposed of hazard-free at the end of their lifetime, potentially unlocking asustainable future for soft robotics.
| |
1_4.png | ONLINE COVER Transplantation Balancing Act. Successful organ transplantation depends on both preventing immune cells from the transplant recipient from rejecting the graft and blocking immune cells from the graft from attacking the host. Zuber et al. demonstrate that a balance in this two-way alloreactivity affects clinical outcomes in human intestinal allograft recipients. [CREDIT: ERAXION/ISTOCKPHOTO.COM] | science | One paradigm in transplantation is that graft-infiltrating T cells are largelynonalloreactive bystander cells. However, the origin and specificity ofallograft T cells over time have not been investigated in detail in animals orhumans. We used polychromatic flow cytometry and high-throughput T cellreceptor sequencing of serial biopsies to show that gut-resident T cellturnover kinetics in human intestinal allografts are correlated with thebalance between intragraft host-versus-graft (HvG) and graft-versus-host (GvH)reactivities and with clinical outcomes. In the absence of rejection, donor Tcells were enriched for GvH-reactive clones that persisted in the long term inthe graft. Early expansion of GvH clones in the graft correlated with therapid replacement of donor antigen-presenting cells by the recipient.Rejection was associated with transient infiltration by blood-like recipientCD28+ NKG2DHi CD8+ T cells, marked predominance of HvG clones, andaccelerated T cell turnover in the graft. Ultimately, these recipient T cellsacquired a steady-state tissue-resident phenotype but regained CD28 expressionduring rejections. Increased ratios of GvH to HvG clones were seen innonrejectors, potentially mitigating the constant threat of rejection posed byHvG clones persisting within the tissue-resident graft T cell population.
| |
2_10.png | ONLINE COVER Promoting Tolerance. Depicted here is a type 1 regulatory T (TR1) cell (left) interacting with a dendritic cell (top right, dark purple) and an interleukin-27 (IL-27)secreting macrophage (bottom right, light purple). Zhang et al. report that IL-27driven commitment of T cells to the TR1 cell lineage limits graft-versus-host disease after bone marrow transplantation. [CREDIT: MADELEINE FLYNN, GRAPHIC SUPPORT OFFICER, QIMR BERGHOFER MEDICAL RESEARCH INSTITUTE AND GARVIN GRULLN, AAAS] | science | Type 1 regulatory T (TR1) cells are Foxp3 interleukin-10 (IL-10)producingCD4+ T cells with potent immunosuppressive properties, but their requirementsfor lineage development have remained elusive. We show that TR1 cellsconstitute the most abundant regulatory population after allogeneic bonemarrow transplantation (BMT), express the transcription factor Eomesodermin(Eomes), and are critical for the prevention of graft-versus-host disease. Wedemonstrate that Eomes is required for TR1 cell differentiation, during whichit acts in concert with the transcription factor B lymphocyteinducedmaturation protein-1 (Blimp-1) by transcriptionally activating IL-10expression and repressing differentiation into other T helper cell lineages.We further show that Eomes induction in TR1 cells requires T-bet and donormacrophagederived IL-27. Thus, we define the cellular and transcriptionalcontrol of TR1 cell differentiation during BMT, opening new avenues totherapeutic manipulation.
| |
2_13.png | ONLINE COVER Teaming Up to Defend. Depicted here are CD141+ dendritic cells (DCs, green), which resist infection by enveloped viruses (yellow spheres), and CD1c+ DCs (orange), which are susceptible to infection. Silvin et al. report that CD141+ DCs acquire viral antigen from infected CD1c+ DCs and prime antiviral T cell responses. [CREDIT: RENAUD CHABRIER, WWW.RENAUDCHABRIR.COM] | science | Dendritic cells (DCs) are critical for the launching of protective T cellimmunity in response to viral infection. Viruses can directly infect DCs,thereby compromising their viability and suppressing their ability to activateimmune responses. How DC function is maintained in light of this paradox isnot understood. By analyzing the susceptibility of primary human DC subsets toviral infections, we report that CD141+ DCs have an innate resistance toinfection by a broad range of enveloped viruses, including HIV and influenzavirus. In contrast, CD1c+ DCs are susceptible to infection, which enablesviral antigen production but impairs their immune functions and survival. Theability of CD141+ DCs to resist infection is conferred by RAB15, a vesicle-trafficking protein constitutively expressed in this DC subset. We show thatCD141+ DCs rely on viral antigens produced in bystander cells to launch cross-presentationdriven T cell responses. By dissociating viral infection fromantigen presentation, this mechanism protects the functional capacity of DCsto launch adaptive immunity against viral infection.
| |
2_14.png | ONLINE COVER Beware of Spitting Camels! Shown is a camel, a zoonotic source for Middle East respiratory syndrome coronavirus (MERS-CoV) infection, which causes a potentially lethal pneumonia that can transfer between individuals. Zhao et al. report that T cell responses in MERS survivors can be detected in the absence of MERS-specific antibody, suggesting that T cells may be useful in detecting mild or subclinical infections. [CREDIT: CINOBY/ISTOCKPHOTO] | science | The Middle East respiratory syndrome coronavirus (MERS-CoV) causes a highlylethal pneumonia. MERS was recently identified as a candidate for vaccinedevelopment, but most efforts focus on antibody responses, which are oftentransient after CoV infections. CoV-specific T cells are generally long-lived,but the virus-specific T cell response has not been addressed in MERSpatients. We obtained peripheral blood mononuclear cells and/or sera from 21MERS survivors. We detected MERS-CoVspecific CD4+ and CD8+ T cell responsesin all MERS survivors and demonstrated functionality by measuring cytokineexpression after peptide stimulation. Neutralizing (PRNT50) antibody titersmeasured in vitro predicted serum protective ability in infected mice andcorrelated with CD4+ but not CD8+ T cell responses; patients with higherPRNT50 and CD4+ T cell responses had longer intensive care unit stays andprolonged virus shedding and required ventilation. Survivors with undetectableMERS-CoVspecific antibody responses mounted CD8+ T cell responses comparablewith those of the whole cohort. There were no correlations between age,disease severity, comorbidities, and virus-specific CD8+ T cell responses. Inconclusion, measurements of MERS-CoVspecific T cell responses may be usefulfor predicting prognosis, monitoring vaccine efficacy, and identifying MERSpatients with mild disease in epidemiological studies and will complementvirus-specific antibody measurements.
| |
2_15.png | ONLINE COVER On the Clock. During pregnancy, the mother's organ systems, including her immune system, adapt to the growing fetus. By carrying out high-dimensional analyses of immune responses in pregnant women, Aghaeepour et al. have charted immune system adaptations during the course of a human pregnancy. This immunological time line is critical for healthy mothers and babies. [CREDIT: JONYA/ISTOCKPHOTO] | science | The maintenance of pregnancy relies on finely tuned immune adaptations. Wedemonstrate that these adaptations are precisely timed, reflecting an immuneclock of pregnancy in women delivering at term. Using mass cytometry, theabundance and functional responses of all major immune cell subsets werequantified in serial blood samples collected throughout pregnancy. Cellsignalingbased Elastic Net, a regularized regression method adapted from theelastic net algorithm, was developed to infer and prospectively validate apredictive model of interrelated immune events that accurately captures thechronology of pregnancy. Model components highlighted existing knowledge andrevealed previously unreported biology, including a critical role for theinterleukin-2dependent STAT5ab signaling pathway in modulating T cellfunction during pregnancy. These findings unravel the precise timing ofimmunological events occurring during a term pregnancy and provide theanalytical framework to identify immunological deviations implicated inpregnancy-related pathologies.
| |
2_16.png | ONLINE COVER Lining Up to Defend. An image of a spleen from a mouse infected with Listeria monocytogenes was stained for CD169+ marginal zone macrophages (red), L. monocytogenes (green), and B cells (blue). Perez et al. show that strategically positioned marginal zone macrophages drive immune responses to L. monocytogenes by controlling pathogen localization in the spleen. [CREDIT: PEREZ ET AL./SCIENCE IMMUNOLOGY] | science | The spleen is an important site for generating protective immune responsesagainst pathogens. After infection, immune cells undergo rapid reorganizationto initiate and maintain localized inflammatory responses; however, themechanisms governing this spatial and temporal cellular reorganization remainunclear. We show that the strategic position of splenic marginal zone CD169+macrophages is vital for rapid initiation of antibacterial responses. Inaddition to controlling initial bacterial growth, CD169+ macrophagesorchestrate a second phase of innate protection by mediating the transport ofbacteria to splenic T cell zones. This compartmentalization of bacteria withinthe spleen was essential for driving the reorganization of innate immune cellsinto hierarchical clusters and for local interferon- production near sites ofbacterial replication foci. Our results show that both phases of theantimicrobial innate immune response were dependent on CD169+ macrophages,and, in their absence, the series of events needed for pathogen clearance andsubsequent survival of the host was disrupted. Our study provides insight intohow lymphoid organ structure and function are related at a fundamental level.
| |
2_17.png | ONLINE COVER Defending the Wall. A colorized scanning electron microscopic image shows the fungus Candida albicans (purple) transitioning from its commensal yeast phase to a pathogenic hyphal form and invading the underlying oral epithelial cell layer (gold). Verma et al. focused on the role of candidalysin, a fungal toxin produced during hyphal invasion, and found that it stimulates production of epithelial factors that in turn trigger IL-17 production from tissue-protective innate lymphocytes. Candidalysin, rather than hyphal formation and tissue damage, may induce the innate immune response. [CREDIT: BERNHARD HUBE/HKI JENA; COLORIZATION, DONNA STOLZ, UNIVERSITY OF PITTSBURGH] | science | Candida albicans is a dimorphic commensal fungus that causes severe oralinfections in immunodeficient patients. Invasion of C. albicans hyphae intooral epithelium is an essential virulence trait. Interleukin-17 (IL-17)signaling is required for both innate and adaptive immunity to C. albicans.During the innate response, IL-17 is produced by T cells and a poorlyunderstood population of innate-acting CD4+ T cell receptor (TCR)+ cells,but only the TCR+ cells expand during acute infection. Confirming the innatenature of these cells, the TCR was not detectably activated during the primaryresponse, as evidenced by Nur77 eGFP mice that report antigen-specificsignaling through the TCR. Rather, the expansion of innate TCR+ cells wasdriven by both intrinsic and extrinsic IL-1R signaling. Unexpectedly, therewas no requirement for CCR6/CCL20-dependent recruitment or prototypical fungalpattern recognition receptors. However, C. albicans mutants that cannotswitch from yeast to hyphae showed impaired TCR+ cell proliferation andIl17a expression. This prompted us to assess the role of candidalysin, ahyphal-associated peptide that damages oral epithelial cells and triggersproduction of inflammatory cytokines including IL-1. Candidalysin-deficientstrains failed to up-regulate Il17a or drive the proliferation of innateTCR+ cells. Moreover, candidalysin signaled synergistically with IL-17,which further augmented the expression of IL-1/ and other cytokines. Thus,IL-17 and C. albicans , via secreted candidalysin, amplify inflammation in aself-reinforcing feed-forward loop. These findings challenge the paradigm thathyphal formation per se is required for the oral innate response anddemonstrate that establishment of IL-1 and IL-17dependent innate immunity isinduced by tissue-damaging hyphae.
| |
2_7.png | ONLINE COVER TAMpering with tumors. Shown is FcRIV expression (blue) on F4/80+ (green) and CD11b+ (red) macrophages in a subcutaneous melanoma tumor. Lehmann et al. show that different subtypes of these tumor-associated macrophages (TAMs) work with therapeutic antibodies depending on tumor location. [CREDIT: LEHMANN ET AL./SCIENCE IMMUNOLOGY] | science | Despite recent advances in activating immune cells to target tumors, thepresence of some immune cells, such as tumor-associated macrophages (TAMs) ortumor-associated neutrophils (TANs), may promote rather than inhibit tumorgrowth. However, it remains unclear how antibody-dependent tumorimmunotherapies, such as cytotoxic or checkpoint control antibodies, affectdifferent TAM or TAN populations, which abundantly express activating Fcreceptors. In this study, we show that the tissue environment determines whichcellular effector pathways are responsible for antibody-dependent tumorimmunotherapy. Although TAMs derived from Ly6Chigh monocytes recruited by theCCL2-CCR2 axis were critical for tumor immunotherapy of skin tumors, thedestruction of lung tumors was CCL2-independent and required the presence ofcolony-stimulating factor 2dependent tissue-resident macrophages. Ourfindings suggest that TAMs may have a dual role not only in promoting tumorgrowth in certain tissue environments on the one hand but also in contributingto tumor cell destruction during antibody-mediated immunotherapy on the otherhand.
| |
2_9.png | ONLINE COVER Combating colitis. Shown is a hematoxylin and eosin-stained section of distal colon tissue. Mehta et al. report that in macrophages the epigenetic reader SP140 maintains transcriptional programs that control intestinal homeostasis. Nonfunctional polymorphisms in SP140 associate with colitis in both mice and humans. [CREDIT: MEHTA ET AL./SCIENCE IMMUNOLOGY] | science | Epigenetic readers that recognize defined posttranslational modifications onhistones have become desirable therapeutic targets for cancer andinflammation. SP140 is one such bromodomain- and plant homeodomain(PHD)containing reader with immune-restricted expression, and single-nucleotide polymorphisms (SNPs) within SP140 associate with Crohns disease(CD). However, the function of SP140 and the consequences of disease-associated SP140 SNPs have remained unclear. We show that SP140 is criticalfor transcriptional programs that uphold the macrophage state. SP140preferentially occupies promoters of silenced, lineage-inappropriate genesbearing the histone modification H3K27me3, such as the HOXA cluster in humanmacrophages, and ensures their repression. Depletion of SP140 in mouse orhuman macrophages resulted in severely compromised microbe-induced activation.We reveal that peripheral blood mononuclear cells (PBMCs) or B cells fromindividuals carrying CD-associated SNPs within SP140 have defective SP140messenger RNA splicing and diminished SP140 protein levels. Moreover, CDpatients carrying SP140 SNPs displayed suppressed innate immune genesignatures in a mixed population of PBMCs that stratified them from other CDpatients. Hematopoietic-specific knockdown of Sp140 in mice resulted inexacerbated dextran sulfate sodium (DSS)induced colitis, and low SP140levels in human CD intestinal biopsies correlated with relatively lowerintestinal innate cytokine levels and improved response to antitumor necrosisfactor (TNF) therapy. Thus, the epigenetic reader SP140 is a key regulator ofmacrophage transcriptional programs for cellular state, and a loss of SP140due to genetic variation contributes to a molecularly defined subset of CDcharacterized by ineffective innate immunity, normally critical for intestinalhomeostasis.
| |
3_19.png | ONLINE COVER Interfering with Zika. The microscopy image shows a section of a human mid-gestation placental chorionic villi that has been stained to visualize localization of actin (red) and cell nuclei (blue). Yockey et al. have studied the role of placental interferon signaling in the context of Zika virus infection using mouse models and human placental tissue. [CREDIT: YOCKEY ET AL./SCIENCE IMMUNOLOGY] | science | Zika virus (ZIKV) infection during pregnancy is associated with adverse fetaloutcomes, including microcephaly, growth restriction, and fetal demise. Type Iinterferons (IFNs) are essential for host resistance against ZIKV, and IFN-/receptor (IFNAR)deficient mice are highly susceptible to ZIKV infection.Severe fetal growth restriction with placental damage and fetal resorption isobserved after ZIKV infection of type I IFN receptor knockout ( Ifnar1 /)dams mated with wild-type sires, resulting in fetuses with functional type IIFN signaling. The role of type I IFNs in limiting or mediating ZIKV diseasewithin this congenital infection model remains unknown. In this study, wechallenged Ifnar1 / dams mated with Ifnar1 +/ sires with ZIKV. Thisbreeding scheme enabled us to examine pregnant dams that carry a mixture offetuses that express ( Ifnar1 +/) or do not express IFNAR ( Ifnar1 /)within the same uterus. Virus replicated to a higher titer in the placenta ofIfnar1 / than within the Ifnar1 +/ concepti. Yet, rather unexpectedly,we found that only Ifnar1 +/ fetuses were resorbed after ZIKV infectionduring early pregnancy, whereas their Ifnar1 / littermates continue todevelop. Analyses of the fetus and placenta revealed that, after ZIKVinfection, IFNAR signaling in the conceptus inhibits development of theplacental labyrinth, resulting in abnormal architecture of the maternal-fetalbarrier. Exposure of midgestation human chorionic villous explants to type IIFN, but not type III IFNs, altered placental morphology and inducedcytoskeletal rearrangements within the villous core. Our results implicatetype I IFNs as a possible mediator of pregnancy complications, includingspontaneous abortions and growth restriction, in the context of congenitalviral infections.
| |
3_20.png | ONLINE COVER Finding the Islets. Shown on the cover is the location of the human pancreas. Culina et al. show that although islet-reactive T cells are present in peripheral circulation in healthy individuals and in type 1 diabetics, these T cells home to the pancreas only in diabetic individuals. Their findings suggest that the ability of these islet-reactive T cells to infiltrate the pancreas is a hallmark of type 1 diabetes. [CREDIT: ISTOCK.COM/NERTHUZ] | science | The human leukocyte antigenA2 (HLA-A2)restricted zinc transporter 8186194(ZnT8186194) and other islet epitopes elicit interferon- secretion by CD8+ Tcells preferentially in type 1 diabetes (T1D) patients compared with controls.We show that clonal ZnT8186194-reactive CD8+ T cells express private T cellreceptors and display equivalent functional properties in T1D and healthyindividuals. Ex vivo analyses further revealed that CD8+ T cells reactive toZnT8186194 and other islet epitopes circulate at similar frequencies andexhibit a predominantly nave phenotype in age-matched T1D and healthy donors.Higher frequencies of ZnT8186194-reactive CD8+ T cells with a more antigen-experienced phenotype were detected in children versus adults, irrespective ofdisease status. Moreover, some ZnT8186194-reactive CD8+ T cell clonotypeswere found to cross-recognize a Bacteroides stercoris mimotope. Whereas ZnT8was poorly expressed in thymic medullary epithelial cells, variable thymicexpression levels of islet antigens did not modulate the peripheral frequencyof their cognate CD8+ T cells. In contrast, ZnT8186194-reactive cells wereenriched in the pancreata of T1D patients versus nondiabetic and type 2diabetic individuals. Thus, islet-reactive CD8+ T cells circulate in mostindividuals but home to the pancreas preferentially in T1D patients. Weconclude that the activation of this common islet-reactive T cell repertoireand progression to T1D likely require defective peripheral immunoregulationand/or a proinflammatory islet microenvironment.
| |
3_22.png | ONLINE COVER Cellular Blackjack. Shown is a dot plot pinpointing the location of interleukin-21(IL-21)producing cells (black dots) among nine clusters of follicular helper T (TFH) cells in lymph nodes from HIV-infected humans. Wendel et al. find that chronic HIV infection is associated with emergence of an excess of activated TFH cells secreting IL-21. [CREDIT: B. S. WENDEL ET AL./SCIENCE IMMUNOLOGY] | science | Follicular helper CD4+ T cells (TFH) play an integral role in promoting B celldifferentiation and affinity maturation. Whereas TFH cell frequencies areincreased in lymph nodes (LNs) from individuals infected with HIV, humoralimmunity remains impaired during chronic HIV infection. Whether HIV inhibitsTFH responses in LNs remains unclear. Advances in this area have been limitedby the difficulty of accessing human lymphoid tissues. Here, we combined high-dimensional mass cytometry with T cell receptor repertoire sequencing tointerrogate the composition of TFH cells in primary human LNs. We foundevidence for intact antigen-driven clonal expansion of TFH cells and selectiveutilization of specific complementarity-determining region 3 (CDR3) motifsduring chronic HIV infection, but the resulting TFH cells acquired anactivation-related TFH cell signature characterized by interleukin-21 (IL-21)dominance. These IL-21+ TFH cells contained an oligoclonal HIV-reactivepopulation that preferentially accumulated in patients with severe HIVinfection and was associated with aberrant B cell distribution in the same LN.These data indicate that TFH cells remain capable of responding to HIVantigens during chronic HIV infection but become functionally skewed andoligoclonally restricted under persistent antigen stimulation.
| |
3_23.png | ONLINE COVER Which Path Gets the Nod? Lymphocyte differentiation involves cells committing to follow one of several diverging pathways. Schmolka et al. show that the capacity of IL-17producing T cells to also secrete IFN- is held in check by expression of the miR-146a microRNA. miR-146a targets mRNA encoding NOD1, a newly appreciated driver of IFN- production. [CREDIT: ISTOCK.COM/FLOCU] | science | T cells are major providers of proinflammatory cytokines. They arepreprogrammed in the mouse thymus into distinct subsets producing eitherinterleukin-17 (IL-17) or interferon- (IFN-), which segregate with CD27expression. In the periphery, CD27 (27) T cells can be induced underinflammatory conditions to coexpress IL-17 and IFN-; the molecular basis ofthis functional plasticity remains to be determined. On the basis ofdifferential microRNA (miRNA) expression analysis and modulation in T cellsubsets, we identified miR-146a as a thymically imprinted post-transcriptionalbrake to limit IFN- expression in 27 T cells in vitro and in vivo. On thebasis of biochemical purification of Argonaute 2bound miR-146a targets, weidentified Nod1 to be a relevant mRNA target that regulates T cellplasticity. In line with this, Nod1 -deficient mice lacked multifunctionalIL-17+ IFN-+ 27 cells and were more susceptible to Listeriamonocytogenes infection. Our studies establish the miR-146a/NOD1 axis as akey determinant of T cell effector functions and plasticity.
| |
3_24.png | ONLINE COVER Defenders in residence. Buggert et al. report that HIV-specific resident memory CD8+ T cells are a central component of antiviral surveillance in HIV-infected lymphoid tissue. Featured on the cover is a confocal microscopic image demonstrating the presence of resident memory CD8+ T cells in a cervical lymph node section from an HIV-infected individual stained with antibodies against CD8 (green), CD69 (red), and CD103 (pink) to identify resident memory CD8+ T cells and CD20 (light blue) to demarcate lymphoid follicles (nuclear staining is shown in blue). [CREDIT: CONSTANTINOS PETROVAS] | science | Current paradigms of CD8+ T cellmediated protection in HIV infection centeralmost exclusively on studies of peripheral blood, which is thought to providea window into immune activity at the predominant sites of viral replication inlymphoid tissues (LTs). Through extensive comparison of blood, thoracic ductlymph (TDL), and LTs in different species, we show that many LT memory CD8+ Tcells bear phenotypic, transcriptional, and epigenetic signatures of residentmemory T cells (TRMs). Unlike their circulating counterparts in blood or TDL,most of the total and follicular HIV-specific CD8+ T cells in LTs alsoresemble TRMs. Moreover, high frequencies of HIV-specific CD8+ TRMs withskewed clonotypic profiles relative to matched blood samples are present inLTs of individuals who spontaneously control HIV replication in the absence ofantiretroviral therapy (elite controllers). Single-cell RNA sequencinganalysis confirmed that HIV-specific TRMs are enriched for effector-relatedimmune genes and signatures compared with HIV-specific non-TRMs in elitecontrollers. Together, these data indicate that previous studies in blood havelargely failed to capture the major component of HIV-specific CD8+ T cellresponses resident within LTs.
| |
3_25.png | ONLINE COVER Binary Decision. Featured on the cover is a photograph of two of the moai statues located on Rapa Nui (Easter Island), a Polynesian island where a Streptomyces bacterial strain was isolated that yielded the macrolide drug rapamycin. Yang et al. report that commitment of T cell precursors to the T cell lineage is compromised in mice with perturbed metabolic signaling caused by absence of the RAPTOR protein, an essential component of the mechanistic target of rapamycin complex 1 (mTORC1). The number two is the organizing theme of this second anniversary issue. [CREDIT: iStock.com/AlbertoLoyo] | science | The interaction between extrinsic factors and intrinsic signal strengthgoverns thymocyte development, but the mechanisms linking them remain elusive.We report that mechanistic target of rapamycin complex 1 (mTORC1) couplesmicroenvironmental cues with metabolic programs to orchestrate the reciprocaldevelopment of two fundamentally distinct T cell lineages, the and Tcells. Developing thymocytes dynamically engage metabolic programs includingglycolysis and oxidative phosphorylation, as well as mTORC1 signaling. Loss ofRAPTOR-mediated mTORC1 activity impairs the development of T cells butpromotes T cell generation, associated with disrupted metabolic remodelingof oxidative and glycolytic metabolism. Mechanistically, we identifymTORC1-dependent control of reactive oxygen species production as a keymetabolic signal in mediating and T cell development, and perturbationof redox homeostasis impinges upon thymocyte fate decisions andmTORC1-associated phenotypes. Furthermore, single-cell RNA sequencing andgenetic dissection reveal that mTORC1 links developmental signals from T cellreceptors and NOTCH to coordinate metabolic activity and signal strength. Ourresults establish mTORC1-driven metabolic signaling as a decisive factor forreciprocal and T cell development and provide insight into metaboliccontrol of cell signaling and fate decisions.
| |
3_27.png | ONLINE COVER Lights, camera, action. Featured on the cover is a hand-drawn sketch of a microscopy room used for intravital imaging by authors Mikael Pittet and Sean Arlauckas. In this Review, Pittet et al. examine the fundamentals of using intravital microscopic imaging to study single cells and highlight how this technology has allowed us to directly visualize the movement and activities of immune cells in various tissues. They also provide a comprehensive overview of mouse reporter lines that are currently available to monitor immune cell dynamics using intravital imaging. [CREDIT: MIKAEL PITTET & SEAN ARLAUCKAS] | science | Intravital microscopic imaging can uncover fundamental aspects of immune cellbehavior in real time in both healthy and pathological states. Here, wediscuss approaches for single-cell imaging of adaptive and innate immune cellsto explore how they migrate, communicate, and mediate regulatory or effectorfunctions in various tissues throughout the body. We further review howintravital single-cell imaging can be used to study drug effects on immunecells.
| |
3_28.png | ONLINE COVER Brush Strokes. The cover features an immunofluorescence image showing three brush cells in mouse tracheal epithelium that co-express the chemosensory cell marker doublecortin-like kinase 1 (DCLK1; green) and the cytokine interleukin-25 (IL-25; red). Airway brush cells display a characteristic tuft of microvilli at their apical tip and express receptors also found in taste-sensing cells. Bankova et al. report that aeroallergen inhalation induces an increase in IL-25producing brush cells that depends on the lipid mediator leukotriene E4 activating the cysteinyl leukotriene 3 receptor. [CREDIT: BANKOVA ET AL./SCIENCE IMMUNOLOGY] | science | Respiratory epithelial cells (EpCs) orchestrate airway mucosal inflammation inresponse to diverse environmental stimuli, but how distinct EpC programs areregulated remains poorly understood. Here, we report that inhalation ofaeroallergens leads to expansion of airway brush cells (BrCs), specializedchemosensory EpCs and the dominant epithelial source of interleukin-25(IL-25). BrC expansion was attenuated in mice lacking either LTC4 synthase,the biosynthetic enzyme required for cysteinyl leukotriene (CysLT) generation,or the EpC receptor for leukotriene E4 (LTE4), CysLT3R. LTE4 inhalation wassufficient to elicit CysLT3R-dependent BrC expansion in the murine airwaythrough an IL-25dependent but STAT6-independent signaling pathway. Last,blockade of IL-25 attenuated both aeroallergen and LTE4-elicitedCysLT3R-dependent type 2 lung inflammation. These results demonstrate thatCysLT3R senses the endogenously generated lipid ligand LTE4 and regulatesairway BrC number and function.
| |
3_29.png | ONLINE COVER Double Feature. This month's cover illustration is an artist's depiction of changes in intratumoral CD8+ T cell function elicited by combination cancer immunotherapy. Wang et al. report that treatment with both agonist anti-GITR and blocking anti-PD-1 antibodies (only affecting the right half of the T cell) triggers transcriptional pathways that lead to enhanced cellular activation (orange cell membrane) and increased cytokine production (dark blue ovals emerging from cell surface). Synergistic activation of the in vivo CD8+ T cell response is associated with robust tumor elimination (with tan cancer cells infiltrating only alongside the untreated left half of the T cell). [CREDIT: RACHEL URKOWITZ] | science | Most patients with cancer do not develop durable antitumor responses afterprogrammed cell death protein 1 (PD-1) or programmed cell death ligand1(PD-L1) checkpoint inhibition monotherapy because of an ephemeral reversal ofT cell dysfunction and failure to promote long-lasting immunological T cellmemory. Activating costimulatory pathways to induce stronger T cell activationmay improve the efficacy of checkpoint inhibition and lead to durableantitumor responses. We performed single-cell RNA sequencing of more than 2000tumor-infiltrating CD8+ T cells in mice receiving both PD-1 and GITR(glucocorticoid-induced tumor necrosis factor receptorrelated protein)antibodies and found that this combination synergistically enhanced theeffector function of expanded CD8+ T cells by restoring the balance of keyhomeostatic regulators CD226 and T cell immunoreceptor with Ig and ITIMdomains (TIGIT), leading to a robust survival benefit. Combination therapydecreased CD8+ T cell dysfunction and induced a highly proliferative precursoreffector memory T cell phenotype in a CD226-dependent manner. PD-1 inhibitionrescued CD226 activity by preventing PD-1Src homology region 2 (SHP2)dephosphophorylation of the CD226 intracellular domain, whereas GITR agonismdecreased TIGIT expression. Unmasking the molecular pathways driving durableantitumor responses will be essential to the development of rationalapproaches to optimizing cancer immunotherapy.
| |
4_31.png | ONLINE COVER Nuclear F-Actin Network in Command. In this issue, Tsopoulidis et al. report that upon T cell activation, nuclear actin rapidly polymerizes into a dense filament network required for CD4+ T cell effector functions including cytokine production. Featured on the cover is a superresolution image prepared using stimulated emission depletion (STED) microscopy that demonstrates the nuclear F-actin network formed 30 seconds after T cell stimulation. [CREDIT: N. TSOPOULIDIS ET AL./SCIENCE IMMUNOLOGY] | science | T cell antigen receptor (TCR) signaling triggers selective cytokine expressionto drive T cell proliferation and differentiation required for immune defenseand surveillance. The nuclear signaling events responsible for specificity incytokine gene expression upon T cell activation are largely unknown. Here, weuncover formation of a dynamic actin filament network in the nucleus thatregulates cytokine expression for effector functions of CD4+ T lymphocytes.TCR engagement triggers the rapid and transient formation of a nuclear actinfilament network via nuclear Arp2/3 complex, induced by elevated nuclear Ca2+levels and regulated via N-Wasp and NIK. Specific interference with TCR-induced formation of nuclear actin filaments impairs production of effectorcytokines and prevents generation of antigen-specific antibodies but does notinterfere with immune synapse formation and cell proliferation. Ca2+-regulatedactin polymerization in the nucleus allows CD4+ T cells the rapid conversionof TCR signals into effector functions required for T cell help.
| |
4_32.png | ONLINE COVER Monitoring Anticommensal Antibodies. This month's cover illustration is a scanning electron micrograph of human feces in which individual commensal bacteria have been colorized to highlight the range of different species present. Paun et al. report development of an assay platform to detect serum anticommensal antibodies that correlate with development of autoimmune and inflammatory diseases. This assay enabled identification of changes in the titer of anticommensal antibodies that exhibit an HLA-dependent association with progression to type 1 diabetes in prediabetic children with islet autoantibodies. [CREDIT: STEVE GSCHMEISSNER/Science Source] | science | Microbiome sequence analyses have suggested that changes in gut bacterialcomposition are associated with autoimmune disease in humans and animalmodels. However, little is known of the mechanisms through which the gutmicrobiota influences autoimmune responses to distant tissues. Here, weevaluated systemic antibody responses against cultured human gut bacterialstrains to determine whether observed patterns of anticommensal antibody(ACAb) responses are associated with type 1 diabetes (T1D) in two cohorts ofpediatric study participants. In the first cohort, ACAb responses in seracollected from participants within 6 months of T1D diagnosis were comparedwith age-matched healthy controls and also with patients with recent onsetCrohns disease. ACAb responses against multiple bacterial speciesdiscriminated among these three groups. In the second cohort, we asked whetherACAb responses present before diagnosis were associated with later T1Ddevelopment and with HLA genotype in participants who were discordant forsubsequent progression to diabetes. Serum IgG2 antibodies against Roseburiafaecis and against a bacterial consortium were associated with future T1Ddiagnosis in an HLA DR3/DR4 haplotypedependent manner. These analysesreveal associations between antibody responses to intestinal microbes andHLA-DR genotype and islet autoantibody specificity and with a futurediagnosis of T1D. Further, we present a platform to investigate antibacterialantibodies in biological fluids that is applicable to studies of autoimmunediseases and responses to therapeutic interventions.
| |
4_33.png | ONLINE COVER Clonality of the Dendritic Cell Network in Tissues. This issue's cover illustration is a Voronoi transformation of a fluorescence microscopy image showing conventional dendritic cell (cDC) clusters in the small intestine of a cDC-specific Confetti mouse. Colors depict distinct fluorescent proteins. Cabeza-Cabrerizo et al. analyze these images to show that cDCs are often organized in tissues as clusters of sister cells that must have divided locally before or after terminal differentiation. Infection can break up these clusters, in part by inducing rapid influx of cDCs precursors that intermingle into the existing mosaic. [CREDIT: MAR CABEZA-CABRERIZO AND DANIEL HEIM] | science | Conventional dendritic cells (cDCs) are found in all tissues and play a keyrole in immune surveillance. They comprise two major subsets, cDC1 and cDC2,both derived from circulating precursors of cDCs (pre-cDCs), which exited thebone marrow. We show that, in the steady-state mouse, pre-cDCs enteringtissues proliferate to give rise to differentiated cDCs, which themselves haveresidual proliferative capacity. We use multicolor fate mapping of cDCprogenitors to show that this results in clones of sister cDCs, most of whichcomprise a single cDC1 or cDC2 subtype, suggestive of pre-cDC commitment. Uponinfection, a surge in the influx of pre-cDCs into the affected tissue dilutesclones and increases cDC numbers. Our results indicate that tissue cDCs can beorganized in a patchwork of closely positioned sister cells of the same subsetwhose coexistence is perturbed by local infection, when the bone marrowprovides additional pre-cDCs to meet increased tissue demand.
| |
4_34.png | ONLINE COVER Superagonist Autoantigen. Featured on the cover is a computer-generated illustration of a 3D protein structure showing the and chains of a human T cell receptor (TCR) recognizing a high-affinity ligand formed by binding of a modified insulin B-chain peptide to the major histocompatibility complex (MHC) class II antigen HLA-DQ8. Wang et al. used x-ray crystallography to study human and mouse TCR recognition of variant forms of an insulin B-chain peptide. C-terminal modifications of this peptide yield peptide-MHC complexes that are potent ligands for autoreactive diabetogenic T cells. These studies provide fresh insights into the structural features of the peptide ligands that spark early steps in human type 1 diabetes. [CREDIT: JOHN KAPPLER] | science | In type 1 diabetes (T1D), proinsulin is a major autoantigen and the insulinB:9-23 peptide contains epitopes for CD4+ T cells in both mice and humans.This peptide requires carboxyl-terminal mutations for uniform binding in theproper position within the mouse IAg7 or human DQ8 major histocompatibilitycomplex (MHC) class II (MHCII) peptide grooves and for strong CD4+ T cellstimulation. Here, we present crystal structures showing how these mutationscontrol CD4+ T cell receptor (TCR) binding to these MHCII-peptide complexes.Our data reveal stricking similarities between mouse and human CD4+ TCRs intheir interactions with these ligands. We also show how fusions betweenfragments of B:9-23 and of proinsulin C-peptide create chimeric peptides withactivities as strong or stronger than the mutated insulin peptides. We proposetranspeptidation in the lysosome as a mechanism that could accomplish thesefusions in vivo, similar to the creation of fused peptide epitopes for MHCIpresentation shown to occur by transpeptidation in the proteasome. Were thismechanism limited to the pancreas and absent in the thymus, it could providean explanation for how diabetogenic T cells escape negative selection duringdevelopment but find their modified target antigens in the pancreas to causeT1D.
| |
4_35.png | ONLINE COVER Adiposphere Mound. This month's cover shows a stack of two 3D organoids grown from distinct subtypes of mesenchymal stromal cells found in visceral adipose tissue. These spherical organoids were generated using adipogenic media and stained with dyes that bound to nuclear DNA (blue) or fat droplets (green). Spallanzani et al. used single-cell RNA sequencing to identify which subtypes of stromal cells within white fat make interleukin-33 (IL-33), a cytokine that supports maintenance of regulatory T cells and group 2 innate lymphoid cells within visceral adipose tissue. [CREDIT: RAUL GERMN SPALLANZANI] | science | Regulatory T cells (Tregs) are key brakes on the visceral adipose tissue (VAT)inflammation that regulates local and systemic metabolic tenor. Breakdown ofthis regulation promotes type 2 diabetes. The cytokine IL-33 expands andsustains the unique Treg population residing within VAT. Here, relying onsingle-cell RNA sequencing, we identified the major IL-33 producers in VAT tobe particular mesenchymal stromal cell subtypes, related to but distinct fromadipocyte progenitor cells. We explored modulation of the VAT stromal celllandscape with physiologic variables such as age and sex, as well as itsremodeling in pathogenic states like obesity. Last, we uncovered a VATTreg:stromal cell negative regulatory loop that keeps the potent effect ofIL-33 under rein.
| |
4_36.png | ONLINE COVER Amoeboid Prowling by Lung ILC2s. Featured on the cover is a confocal image of a precision-cut mouse lung slice showing green group 2 innate lymphoid cells (ILC2s) in close proximity to peribronchial collagen fibers (in blue and magenta). Puttur et al. used intravital microscopy techniques to track the migration of ILC2s during IL-33induced lung inflammation. These studies identify type I collagen and chemokines as key influences on the shape and locomotor activity of ILC2s during lung inflammation. [CREDIT: FRANZ PUTTUR ET AL./SCIENCE IMMUNOLOGY] | science | Group 2 innate lymphoid cells (ILC2s) are enriched in mucosal tissues (e.g.,lung) and respond to epithelial cellderived cytokines initiating type 2inflammation. During inflammation, ILC2 numbers are increased in the lung.However, the mechanisms controlling ILC2 trafficking and motility withininflamed lungs remain unclear and are crucial for understanding ILC2 functionin pulmonary immunity. Using several approaches, including lung intravitalmicroscopy, we demonstrate that pulmonary ILC2s are highly dynamic, exhibitamoeboid-like movement, and aggregate in the lung peribronchial andperivascular spaces. They express distinct chemokine receptors, includingCCR8, and actively home to CCL8 deposits located around the airway epithelium.Within lung tissue, ILC2s were particularly motile in extracellularmatrixenriched regions. We show that collagen-I drives ILC2 to markedlychange their morphology by remodeling their actin cytoskeleton to promoteenvironmental exploration critical for regulating eosinophilic inflammation.Our study provides previously unappreciated insights into ILC2 migratorypatterns during inflammation and highlights the importance of environmentalguidance cues in the lung in controlling ILC2 dynamics.
|
End of preview. Expand
in Dataset Viewer.
Dataset Card for MAC
The Multimodal Academic Cover (MAC) is a benchmark, comprising a 5872 collection of cover images, cover stories, and relevant articles from leading academic journals, including Cell, Nature, Science, and their sub-publications. MAC is designed to test the ability of Multimodal models on the scientific visual understanding.
Dataset Details
MAC consists of 5872 journal issues, each with a complete group of cover images, cover stories, and articles. Two subsets are also provided, MAC-Medium (940 issues) and MAC-Small (50 issues), to facilitate fast or qualitative evaluation.
- Curated by: Jin Gao, Jiahao Zhan, Chongxuan Li, Dequan Wang
- Language(s) (NLP): [English]
Uses
from datasets import load_dataset
dataset = load_dataset("JohnZhan/MAC","MAC-Small",split="train")
print(dataset["train"][0])
- Downloads last month
- 0