Myungkyu/banana-challenge-data

banana challenge — LeRobot dataset

Teleoperated demonstrations of "Place the banana into the basket" on an OpenArm + Inspire Hand platform (right arm and hand only).

task	Place the banana into the basket (single instruction)
episodes / frames	54 / 16,202
fps	30
cameras	ego_left, ego_right — 640×360 RGB, stereo ego view
state / action	13-D absolute joint: 7 arm (openarm_right_joint1..7) + 6 hand (right_hand_{thumb_1,thumb_2,index_1,middle_1,ring_1,little_1}_joint)

Layout

Two LeRobot generations of the same recording, built from the same raw bags:

• v2.1/ — LeRobot v2.1. Carries meta/modality.json, episodes.jsonl, stats.json. Consumed by RLDX-1 / GR00T-style stacks, which read v2.1 only.
• v3.0/ — LeRobot v3.0, derived from v2.1/ with the official converter and quantile statistics recomputed. Consumed by lerobot's pi0.5 training.

Frames, videos and joint values are identical across the two; only the metadata format differs.

Notes

• Actions are absolute joint targets, not deltas.
• Normalization uses q01/q99 quantile statistics, present in both generations' stats.json. The official v2.1→v3.0 converter drops them, so they were recomputed after conversion.
• The head/neck joints were held fixed during collection (openarm_head_pitch ≈ 1.0 rad, openarm_head_yaw = 0.0 rad) and are not part of the 13-D vector. The left arm was not used.

Models trained on this dataset: Myungkyu/banana-challenge-ckpts