video video 14.5 28.4 |
|---|
CoLA three-arm handover chain
Three ALOHA arms pass a red box along a chain, A -> B -> C.
- 232 episodes: every successful episode from a 300-attempt scripted collection (232/300 succeeded; failures were discarded).
- Episode length 218-427 steps (mean 228.6).
- The paper uses 130 of them (104 / 13 / 13), listed in
splits.json; the remaining 102 are extra successful demonstrations not used in the paper.
Format
One HDF5 file per episode (episode_NNNN.h5), recorded at 50 Hz control
(MuJoCo, timestep 2 ms x 10 substeps), plus a rendered review video per episode
(episode_NNNN_review.mp4). Datasets inside each .h5, T = episode length:
| key | shape | dtype | meaning |
|---|---|---|---|
image_wrist_a |
(T, 256, 256, 3) | uint8 | arm A wrist camera |
image_wrist_b |
(T, 256, 256, 3) | uint8 | arm B wrist camera |
image_wrist_c |
(T, 256, 256, 3) | uint8 | arm C wrist camera |
image_overhead |
(T, 256, 256, 3) | uint8 | fixed overhead camera |
state_a |
(T, 7) | float64 | arm A joint state |
action_a |
(T, 7) | float64 | arm A action |
state_b |
(T, 7) | float64 | arm B joint state |
action_b |
(T, 7) | float64 | arm B action |
state_c |
(T, 7) | float64 | arm C joint state |
action_c |
(T, 7) | float64 | arm C action |
box_pos |
(T, 3) | float64 | box position (m) |
phase |
(T,) | bytes | scripted-expert phase label |
state_* is the arm's 6 joint positions plus finger position; action_* is
the 6 absolute joint-position targets plus the gripper command. Images are
256x256 RGB. File attributes: success, handover_only, policy_module,
scan_found.
Every episode starts with arm A already holding the box and was produced by a scripted expert; only successful episodes are included.
Splits
splits.json gives the train/val/test split used in the paper (80/10/10,
seed 42, stratified by success). Each entry is a filename in this repo.
- Downloads last month
- 237