MikelRubi/act-pick
Robotics • Updated
Datasets:
350 episodes · 30 fps · 2 cameras · 640×480 h264
robotekin-so101-pick
Pick the bracket from the tray and hold it up to the inspection camera. Stage 1 · Pick of the SO-101 pin-sorting cell built at the Robotekin LeRobot Hackathon 2026.
Project page: MikelRubi/robotekin-so101-pin-sorting. Trained model: MikelRubi/act-pick.
Naming note:
good/baddescribe the part, not the quality of a demonstration. GOOD = part with a pin, BAD = part without a pin (defective). In the usual order GOOD parts go to the green box and BAD parts to the white box. Every episode is a successful demonstration; no failed or negative demonstrations were intentionally recorded.
What it contains
The arm picks one metal bracket from the orange tray, grasping it according to its orientation, lifts it and holds it up in front of the inspection camera. Each episode ends with the part held there, ready for the pin classifier. The stored task is "Pick a bracket from the crate and hold it up to the camera".
| Episodes | 350 (93,206 frames, 51.8 min) |
| Sessions joined | 41 (listed below and in meta/sources.json) |
| Frame rate | 30 fps |
| Cameras | observation.images.front, observation.images.wrist (640×480 RGB, H.264) |
| State / action | observation.state / action, 6 values each |
Usage
from lerobot.datasets.lerobot_dataset import LeRobotDataset
ds = LeRobotDataset("MikelRubi/robotekin-so101-pick")
How it was used
Source of MikelRubi/act-pick (v1–v3; each version filters episodes differently, see the model card). Two sessions were held out for evaluation: so101-pick-cube_20261024_181939 and so101-pick-cube_20261024_144537. The 23 October warm-up sessions, made with the first recorder, are not included here.
Robot and recording
- Robot: SO-101 follower arm (LeRobot
so101_follower), teleoperated with an SO-101 leader arm.actionis the leader position sent to the follower;observation.stateis the measured follower position. - Joints (same order in state and action):
shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper. Arm joints in degrees, gripper 0–100. - Cameras:
front(fixed overhead camera looking down at the tray; the part is lifted towards it for inspection) andwrist(on the gripper), RGB 640×480 at 30 fps. - Recorded on 24 October 2026 at the Robotekin LeRobot Hackathon (Vitoria-Gasteiz) with the team's recorder
(
17b_record_real_so101_live.py), which keeps only the takes the operator accepts. Each session was uploaded as its own dataset; this one joins them with LeRobot'saggregate_datasets(LeRobot 0.6.1, dataset format v3.0).
Source sessions (in episode order)
Limitations
- One room and lighting setup, recorded on one day; no success labels are stored (every saved episode was accepted by the operator).
- Camera models and their exact mounting positions are not documented.
Acknowledgements
Recorded by Iker Azurmendi, Mikel Zulueta, Ander Autillo and Diego Suarez during the Robotekin LeRobot Hackathon 2026 (23–25 October 2026, Vitoria-Gasteiz). Thanks to Bravo Robots and to Robotekin for the hackathon, to Hugging Face for LeRobot, and to everyone who took part in it.
- Downloads last month
- 133
Models trained or fine-tuned on MikelRubi/robotekin-so101-pick