Bi-Rokae real-robot demonstrations, 6 tasks (301 episodes)
Bimanual Rokae AR5 arms with parallel grippers. Companion to Noctis12138/egoverse-aria-selected
(the ego episodes were retrieved using these 6 tasks). Converted from the original LeRobot v3.0 recordings to
LeRobot v2.1 (openpi's pinned lerobot). 301 episodes, 192,158 frames
(1.78 h) at 30 Hz, 5.3 GB. Private. Do not redistribute.
| source set | episodes | frames | prompt (task) |
|---|---|---|---|
| dual_arm_block_cup_placement | 51 | 28,749 | place the block into the cup with the left arm, then place the cup onto the plate with the right arm |
| dual_arm_cross_layer_swap | 50 | 46,052 | Use the left hand to place the red block on the higher shelf, and use the right hand to place the orange block on the lower shelf. |
| dual_arm_cup_stack_and_place | 50 | 38,993 | stack one cup into the other, then place the stacked cups onto the plate |
| dual_arm_drawer_block_place | 50 | 24,863 | open the drawer, put the pink block into the drawer, then close the drawer |
| dual_arm_toothbrush_in_cup_to_yellow_shelf | 50 | 21,674 | place the cup on the yellow shelf with the left arm, then put the toothbrush into the cup with the right arm |
| pomegranate_orange_to_purple_box | 50 | 31,827 | put the orange and the pomegranate into the purple box |
Prompts: in the source tasks.parquet, four sets (block_cup, cup_stack, drawer_block, pomegranate) contain the
placeholder "待定" and toothbrush carries a copy of the cross_layer sentence; only cross_layer has its own prompt, kept
as is. The other five prompts were written from the head-camera video and approved by the project lead; they are not
the data owner's wording. Changing them only needs meta/tasks.jsonl and meta/episodes.jsonl (one task per set, the
task_index in the parquet files stays valid).
# openpi pins lerobot 0cf8648 (codebase v2.1); the repo carries the `v2.1` tag the loader asks for.
# Private repo: `huggingface-cli login` (or HF_TOKEN) with an account that has access.
from lerobot.common.datasets.lerobot_dataset import LeRobotDataset
ds = LeRobotDataset("Noctis12138/rokae_bimanual_6task", delta_timestamps={"action": [t / 30 for t in range(50)]})
item = ds[0] # item["action"]: (50, 16) absolute EE poses; item["task"]: prompt string
Features
| key | shape | content |
|---|---|---|
| observation.images.head / left_wrist / right_wrist | 480×640×3 video (AV1) | the three original cameras, re-encoded |
| observation.state | 16 | measured EE poses, layout below |
| action | 16 | commanded EE poses (from the original action), same layout |
| loss_mask | 16 | all ones |
| observation.state_raw / action_raw | 30 | the original 30-d vectors, names preserved (7 joints, cart_pos0..5, psi, gripper per arm) |
16-d EE layout
| dims | meaning |
|---|---|
| 0-2 | left EE position x, y, z (m) |
| 3-6 | left EE orientation quaternion (x, y, z, w), sign made continuous over time |
| 7 | left gripper |
| 8-10 | right EE position x, y, z (m) |
| 11-14 | right EE orientation quaternion (x, y, z, w) |
| 15 | right gripper |
Definitions (robot side)
- Position:
cart_pos0..2(m). Each arm is expressed in its own base frame (left and right are not in one frame). - Orientation:
cart_pos3..5read as ROS roll-pitch-yaw (fixed-axis x-y-z,scipy'xyz'), checked earlier against forward kinematics to 2–5°. TCP is 0.092 m along link-7 z (URDF). Tool axes against base z: y ≈ −0.93 (left) / +0.90 (right), z ≈ −0.37 (both). - Gripper:
gripper_pos, 0..1, mostly binary in the data. - Commanded vs measured position differ by 0.8–4.7 mm (median per task and arm).
- Video frames were cut from the v3 concatenated files by timestamp. Episodes where a camera had fewer frames than rows (padded with the last frame): 0.
Notes for pi0.5 co-training
observation.stateandactionhold absolute poses. Chunk-relative actions (pose at t0+k expressed in the t0 frame) are meant to be computed in the training transform, because the horizon H is a training choice.loss_maskhas the same 16-d layout: multiply the per-dimension action loss by it (for a chunk, use the mask of each target frame).- Frame rate is the native 30 Hz for both datasets; nothing was resampled.
- State is included, but the co-training plan does not feed it to the model.
pi0.5 training recipe (ego + robot co-training, and the robot-only baseline)
Values marked [set] were fixed by the project lead; [plan] come from the agreed co-training plan;
[openpi] are openpi defaults / its own pi0.5 fine-tuning configs (openpi main, checked 2026-09-26) and may be tuned.
Run the robot-only baseline with the same settings (only the ego share changes).
| item | value | source |
|---|---|---|
| init weights | gs://openpi-assets/checkpoints/pi05_base/params (full pi0.5 base) |
[plan] |
| batch size | 32 | [set] |
| batch composition | ego : robot = 1 : 1 in every batch (16 + 16); robot-only baseline = 32 robot | [set] |
| learning rate | constant after warmup: CosineDecaySchedule(warmup_steps=1_000, peak_lr=5e-5, decay_steps=30_000, decay_lr=5e-5) (openpi has no constant schedule class; pi05_libero uses this peak = end pattern) |
[set] constant; 5e-5 [openpi pi05_libero]; warmup 1k [openpi default] |
| optimizer | AdamW b1 0.9, b2 0.95, eps 1e-8, weight decay 1e-10, grad-norm clip 1.0 | [openpi] |
| EMA | 0.999 | [openpi pi05_libero] |
| train steps | 30,000 | [openpi default] |
| model | Pi0Config(pi05=True, action_horizon=50, discrete_state_input=False), action_dim 32 (14 used, rest zero-padded), max_token_len 200 |
[openpi] horizon/dims; no state [plan] |
| action | 14-d chunk-relative per arm: translation and rotation (rotvec) of the pose at t0+k expressed in the EE frame at t0, gripper absolute; computed from the absolute 16-d action column |
[plan] |
| state input | none for both embodiments | [plan] |
| normalisation | 1% / 99% quantiles to [-1, 1], per embodiment; ego stats exclude masked dims / frames; de-normalise with robot stats at inference | [plan] |
| loss mask | per-dim flow loss × loss_mask of each target frame (ego gripper dims and untracked hands = 0); drop windows touching frame_valid = 0 |
[plan] |
| discrete (FAST) loss | none for ego samples (openpi pi0.5 fine-tuning trains the flow loss only; if a FAST/discrete auxiliary is added, turn it off for ego) | [plan] |
| cameras | head → base image; left/right wrist → wrist slots; ego wrist slots image_mask = False | [plan] |
| wrist dropout (robot) | 15% per sample, drop one of the two wrist cameras at random; 0 for the last 10–20% of steps; all on at inference; same in baseline | [plan] |
| image augmentation | random crop + colour jitter (both embodiments) | [plan] |
| prompt | task column |
[plan] |
| seed | 42 | [openpi default] |
Conversion script: conversion/c10_convert.py in this repo.
- Downloads last month
- 478