# RoboChallenge Table30 v2 Dataset ## Tasks and Embodiments The dataset includes 30 diverse manipulation tasks (Table30 v2) across 4 embodiments: ### Available Tasks - `put_the_books_back` - Place the books back onto the bookshelf. - `tie_a_knot` - Tie a knot with the string on the table. - `stamp_positioning` - Stamp the signature area on the paper. - `tidy_up_the_makeup_table` - Sort and organize the cosmetics on the table. - `paint_jam` - Spread the bread with jam. - `pack_the_items` - Box up the tablet and its accessories. - `wrap_with_a_soft_cloth` - Bundle the objects together using the cloth on the table. - `put_in_pen_container` - Put the pens on the desk into the pen holder. - `put_the_pencil_case_into_the_schoolbag` - Put the pencil case into the backpack. - `put_the_shoes_back` - Pair the two pairs of shoes on the desk and place them on the shoe rack - `untie_the_shoelaces` - Remove the laces from the shoes, then place them on the table. - `scoop_with_a_small_spoon` - Scoop beans and place them into the empty bowl. - `wipe_the_blackboard` - Wipe the balckboard clean. - `lint_roller_remove_dirt` - Use a lint remover to clean the debris on the clothe. - `turn_on_the_light_switch` - Turn on the lamp. - `hold_the_tray_with_both_hands` - Place the ball on the desk onto the small tray, and then move it to the large tray. - `fold_the_clothes` - Fold the T-shirts and stack them neatly in the upper-left corner of the table. - `pack_the_toothbrush_holder` - Put the toothbrush and toothpaste into the toiletries case in sequence, close the case, and then place it into the basket. - `place_objects_into_desk_drawer` - Open the drawer, put the bottle opener inside, and close the drawer. - `sweep_the_trash` - Sweep the trash on the table into the dustpan. - `arrange_flowers` - Put the 4 flowers into the vase. - `press_the_button` - Press the buttons in the following sequence: pink, blue, green, and then yellow. - `pick_out_the_green_blocks` - Find all the green blocks and put them into the basket. - `hang_the_cup` - Hang the cup on the rack. - `water_the_flowers` - Water the potted plants. - `wipe_the_table` - Wipe the stains off the desk with a rag. - `arrange_fruits` - Arrange the fruit in the basket. - `shred_paper` - Put the paper into the shredder. - `item_classification` - Place the stationery in the yellow box and the electronics in the blue box. - `stack_bowls` - Put the blue bowl into the beige bowl, and put the green bowl into the blue bowl. ### Embodiments - **ARX5** - Single-arm with triple camera setup (wrist + global + right-side views) - **UR5** - Single-arm with dual camera setup (wrist + global views) - **ALOHA** - Dual-arm with triple wrist camera setup (left wrist + right wrist + global views) - **DOS-W1** - Dual-arm with triple wrist camera setup (left wrist + right wrist + global views) ## Dataset Structure ### Hierarchy The dataset is organized by tasks, with each task containing multiple demonstration episodes: ``` . ├── / # e.g., arrange_the_flowers, fold_t_shirt │ ├── task_desc.json # Task description │ ├── meta/ # Task-level metadata │ │ ├── task_info.json │ └── data/ # Episode data │ ├── episode_000000/ # Individual episode │ │ ├── meta/ │ │ │ └── episode_meta.json # Episode metadata │ │ ├── states/ │ │ │ # for single-arm (ARX5, UR5) │ │ │ ├── states.jsonl # Single-arm robot states │ │ │ # for dual-arm (ALOHA, DOS-W1) │ │ │ ├── left_states.jsonl # Left arm states │ │ │ └── right_states.jsonl # Right arm states │ │ └── videos/ │ │ # Video configurations varies by robot model: │ │ # ARX5 │ │ ├── cam_arm_rgb.mp4 # Wrist view │ │ ├── cam_global_rgb.mp4 # Global view │ │ └── cam_side_rgb.mp4 # Side view │ │ # UR5 │ │ ├── cam_global_rgb.mp4 # Global view │ │ └── cam_arm_rgb.mp4 # Wrist view │ │ # ALOHA │ │ ├── cam_high_rgb.mp4 # Global view │ │ ├── cam_left_wrist_rgb.mp4 # Left wrist view │ │ └── cam_right_wrist_rgb.mp4 # Right wrist view │ │ # DOS-W1 │ │ ├── cam_high_rgb.mp4 # Global view │ │ ├── cam_left_wrist_rgb.mp4 # Left wrist view │ │ └── cam_right_wrist_rgb.mp4 # Right wrist view │ ├── episode_000001/ │ └── ... ├── convert_to_lerobot.py # Conversion script └── README.md ``` ### Metadata Schema `task_info.json` ```json { "task_desc": { "task_name": "arrange_flowers", // Task identifier "prompt": "Put the 4 flowers into the vase.", "description": "...", "scoring": "...", // Scoring criteria "task_tag": [ // Task characteristics "repeated", "single-arm", "ARX5", "precise3d" ] }, "video_info": { "fps": 30, // Video frame rate "ext": "mp4", // Video format "encoding": { "vcodec": "libx264", // Video codec "pix_fmt": "yuv420p" // Pixel format } } } ``` `episode_meta.json` ```json { "start_time": 1750405586.3430033, // Unix timestamp (start) "end_time": 1750405642.5247612, // Unix timestamp (end) "frames": 1672, // Total video frames "robot_id": "rc_arx5_5", // Robot identifier "features": { "cam_global": { // Camera name info "intrinsics": [], // Intrinsics "extrinsics": { // Extrinsics "arms": { "arm": [] // Extrinsic relative to arm } } } } } ``` ### Robot States Schema Each episode contains states data stored in JSONL format. Depending on the embodiment, the structure differs slightly: - **Single-arm robots (ARX5, UR5)** → `states.jsonl` - **Dual-arm robots (ALOHA, DOS-W1)** → `left_states.jsonl` and `right_states.jsonl` Each file records the robot’s proprioceptive signals per frame, including joint angles, end-effector poses, gripper states, and timestamps. The exact field definitions and coordinate conventions vary by platform, as summarized below. #### ARX5 | Data Name | Data Key | Shape | Semantics | |:----------------:|:----------------:|:-----:|:---------------------------------------------------------------------------------------------------------------------------------------------------------------:| | Joint control | joint_positions | (6,) | Joint angle (in radians) from the base to the end effector. | | Joint velocity | joint_velocities | (6,) | Speed of 6 joint. | | Joint effort | efforts | (7,) | Effort of 6 joints and gripper. (Provided by official API, precision not guaranteed. | | Pose control | ee_positions | (7,) | End effector pose (tx, ty, tz, rx, ry, rz, rw), where (tx, ty, tz) is relative position from the arm base coordinate , (rx, ry, rz, rw) is quaternion rotation. | | Gripper control | gripper_width | (1,) | Actual gripper width measurement in meter. | | Gripper velocity | gripper_velocity | (1,) | Speed of gripper. | | Time stamp | timestamp | (1,) | Floating point timestamp (in milliseconds) of each frame. | #### UR5 | Data Name | Data Key | Shape | Semantics | |:---------------:|:---------------:|:-----:|:---------------------------------------------------------------------------------------------------------------------------------------------------------------:| | Joint control | joint_positions | (6,) | Joint angle (in radians) from the base to the end effector. | | Pose control | ee_positions | (7,) | End effector pose (tx, ty, tz, rx, ry, rz, rw), where (tx, ty, tz) is relative position from the arm base coordinate , (rx, ry, rz, rw) is quaternion rotation. | | Gripper control | gripper_width | (1,) | Actual gripper width measurement in meter. | | Time stamp | timestamp | (1,) | Floating point timestamp (in milliseconds) of each frame. | #### DOS-W1 | Data Name | Data Key | Shape | Semantics | |:---------------:|:---------------:|:-----:|:---------------------------------------------------------------------------------------------------------------------------------------------------------------:| | Joint control | joint_positions | (6,) | Joint angle (in radians) from the base to the end effector. | | Pose control | ee_positions | (7,) | End effector pose (tx, ty, tz, rx, ry, rz, rw), where (tx, ty, tz) is relative position from the arm base coordinate , (rx, ry, rz, rw) is quaternion rotation. | | Gripper control | gripper_width | (1,) | Actual gripper width measurement in meter. | | Time stamp | timestamp | (1,) | Floating point timestamp (in milliseconds) of each frame. | #### ALOHA | Data Name | Data Key | Shape | Semantics | |:--------------------:|:----------------:|:-----:|:---------------------------------------------------------------------------------------------------------------------------------------------------------------:| | Joint control | joint_positions | (6,) | Puppet joint angle (in radians) from the base to the end effector. | | Joint velocity | joint_velocities | (7,) | Speed of 6 joint. | | Gripper control | gripper_width | (1,) | Actual gripper width measurement in meter. | | Pose control | ee_positions | (7,) | End effector pose (tx, ty, tz, rx, ry, rz, rw), where (tx, ty, tz) is relative position from the arm base coordinate , (rx, ry, rz, rw) is quaternion rotation. | | Joint effort | efforts | (7,) | Effort of 6 joints and gripper. (Provided by official API, precision not guaranteed. | | Time stamp | timestamp | (1,) | Floating point timestamp (in mileseconds) of each frame. | ## Convert to LeRobot While you can implement a custom Dataset class to read RoboChallenge data directly, **we strongly recommend converting to LeRobot format** to take advantage of [LeRobot](https://github.com/huggingface/lerobot)'s comprehensive data processing and loading utilities. The example script **`convert_to_lerobot.py`** converts **ARX5** data to the LeRobot dataset as a example. For other robot embodiments (UR5, ALOHA, DOS-W1), you can adapt the script accordingly. ### Prerequisites - Python 3.9+ with the following packages: - `lerobot@0cf864870cf29f4738d3ade893e6fd13fbd7cdb5` - `opencv-python` - `numpy` - Configure `$HF_LEROBOT_HOME` (defaults to `~/.cache/huggingface/lerobot` if unset). ```bash pip install git+https://github.com/huggingface/lerobot@0cf864870cf29f4738d3ade893e6fd13fbd7cdb5 opencv-python numpy export HF_LEROBOT_HOME="/path/to/lerobot_home" ``` ### Usage Run the converter from the repository root (or provide an absolute path): ```bash python convert_to_lerobot.py \ --repo-name example_repo \ --raw-dataset /path/to/example_dataset \ --frame-interval 1 ``` ### Output - Frames and metadata are saved to `$HF_LEROBOT_HOME/`. - At the end, the script calls `dataset.consolidate(run_compute_stats=False)`. If you require aggregated statistics, run it with `run_compute_stats=True` or execute a separate stats job.