carlogrigioni/safe-road-crossing-aw-dataset

safe_roadcrossing_aw

Dataset for Safe Road Crossing Decision for Autonomous Wheelchairs.

Data collection methodology

The dataset deals with a road-crossing scenario, using multiple sensors. We use laboratory wheeled robots (RoboMaster EP), a motion tracker (Optitrack), cameras and range sensors. The reference framework is defined by the European project REXASI-PRO (REliable & eXplAinable Swarm Intelligence for People with Reduced mObility https://rexasi-pro.spindoxlabs.com/) which addresses indoor and outdoor use cases to demonstrate trustworthy social navigation of autonomous wheelchairs together with flying drones in real-world environments. To this aim, we use laboratory 3 wheeled ground robots, equipped with artificial vision and distance sensors, as mock-up models of actual AWs and drones.

This dataset was collected inside the Mobile Robotics Laboratory of IDSIA https://idsia-robotics.github.io/. The available data is recorded from:

- Robots
  - Three wheeled omni-directional robots: each one is a RoboMaster EP (RM) https://www.dji.com/ch/robomaster-ep, a commercial education platform from DJI. Each RM is customised for its role as “car”, “AW”, or “drone”.
  - Each RM is equipped with reflective tape markers to allow detection from motion tracker.
  - RM car: simulates a vehicle, it is equipped with no additional sensors.
  - RM wheelchair: simulates an Autonomous Wheelchiar, it is equipped with a camera and four infrared range sensors.
  - RM drone: simulates a drone, it is equipped with a camera.
- Motion tracker
  - OptiTrack motion tracker
  - 18 infrared cameras

Description of data

The dataset obtained from experiments in the Mobile Robotics Laboratory is made of 15 recordings. Recordings from 0 to 5 are in Scene A, and recordings from 6 to 14 are in Scene B.

Scene A

Scene B

The dataset is in the folder safe-road-crossing-aw-dataset/. It is composed by 15 folders experiment_{i}/, which cointain all data relative to each experiment.

safe-road-crossing-aw-dataset/
  -experiment_{i}/
    -drone/
      - frame_{000i}.png
      - ...
    -wheelchair/
      - frame_{000i}.png
      - ...
    -bbox_{i}.csv
    -bbox_{i}_drone.csv
    -camera_stamps{i}.csv
    -range_{i}.csv
    -tracker{i}.csv
  -.../
  -README.md

• drone and wheelchair folders: numbered frames extracted from 30 fps, 1280 x 720 videos, recorded from cameras equipped on RM drone and RM wheelchair.

• bbox_{i}.csv and bbox_{i}_drone.csv: bounding-boxes width in pixels extracted by YOLO, with added adverse weather conditions, produced with Automold Road Augmentation library (https://github.com/ UjjwalSaxena/Automold--Road-Augmentation-Library)

  Data in .csv file is organized as:

  Frame	bbox	bbox_bright	bbox_rain	bbox_fog
  png frame number	normal conditions	added light	added rain	added fog

• camera_stamps{i}.csv: absolute timestamps in nanoseconds corresponding to each frame collected from cameras as returned by ROS2

  stamp_w	stamp_d
  wheelchair frame timestamp	drone frame timestamp

• range_{i}.csv: distance in meters of the nearest object collected from infrared distance sensors, absolute timestamps are in nanoseconds.

  timestamp	range_0	range_1	range_2	range_3
  timestamp	distance	distance	distance	distance

• tracker{i}.csv: relative distance in meters of the obstacle and wheelchair calculated from data from the motion tracker, absolute timestamps are in nanoseconds.

  timestamp	distance
  timestamp	distance

File formats

- tabular data, format csv
- images, format png

Acknowledgments

This work was supported by the Swiss State Secretariat for Education, Research and lnnovation (SERI). The project has been selected within the European Union’s Horizon Europe research and innovation programme under grant agreement: HORIZON-CL4-2021-HUMAN-01-01. Views and opinions expressed are however those of the authors only and do not necessarily reflect those of the funding agencies, which cannot be held responsible for them.