Upload README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,234 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# RoboChallenge Table30 v2 Dataset
|
| 2 |
+
|
| 3 |
+
## Tasks and Embodiments
|
| 4 |
+
|
| 5 |
+
The dataset includes 30 diverse manipulation tasks (Table30 v2) across 4 embodiments:
|
| 6 |
+
|
| 7 |
+
### Available Tasks
|
| 8 |
+
|
| 9 |
+
- `put_the_books_back` - Place the books back onto the bookshelf.
|
| 10 |
+
- `tie_a_knot` - Tie a knot with the string on the table.
|
| 11 |
+
- `stamp_positioning` - Stamp the signature area on the paper.
|
| 12 |
+
- `tidy_up_the_makeup_table` - Sort and organize the cosmetics on the table.
|
| 13 |
+
- `paint_jam` - Spread the bread with jam.
|
| 14 |
+
- `pack_the_items` - Box up the tablet and its accessories.
|
| 15 |
+
- `wrap_with_a_soft_cloth` - Bundle the objects together using the cloth on the table.
|
| 16 |
+
- `put_in_pen_container` - Put the pens on the desk into the pen holder.
|
| 17 |
+
- `put_the_pencil_case_into_the_schoolbag` - Put the pencil case into the backpack.
|
| 18 |
+
- `put_the_shoes_back` - Pair the two pairs of shoes on the desk and place them on the shoe rack
|
| 19 |
+
- `untie_the_shoelaces` - Remove the laces from the shoes, then place them on the table.
|
| 20 |
+
- `scoop_with_a_small_spoon` - Scoop beans and place them into the empty bowl.
|
| 21 |
+
- `wipe_the_blackboard` - Wipe the balckboard clean.
|
| 22 |
+
- `lint_roller_remove_dirt` - Use a lint remover to clean the debris on the clothe.
|
| 23 |
+
- `turn_on_the_light_switch` - Turn on the lamp.
|
| 24 |
+
- `hold_the_tray_with_both_hands` - Place the ball on the desk onto the small tray, and then move it to the large tray.
|
| 25 |
+
- `fold_the_clothes` - Fold the T-shirts and stack them neatly in the upper-left corner of the table.
|
| 26 |
+
- `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.
|
| 27 |
+
- `place_objects_into_desk_drawer` - Open the drawer, put the bottle opener inside, and close the drawer.
|
| 28 |
+
- `sweep_the_trash` - Sweep the trash on the table into the dustpan.
|
| 29 |
+
- `arrange_flowers` - Put the 4 flowers into the vase.
|
| 30 |
+
- `press_the_button` - Press the buttons in the following sequence: pink, blue, green, and then yellow.
|
| 31 |
+
- `pick_out_the_green_blocks` - Find all the green blocks and put them into the basket.
|
| 32 |
+
- `hang_the_cup` - Hang the cup on the rack.
|
| 33 |
+
- `water_the_flowers` - Water the potted plants.
|
| 34 |
+
- `wipe_the_table` - Wipe the stains off the desk with a rag.
|
| 35 |
+
- `arrange_fruits` - Arrange the fruit in the basket.
|
| 36 |
+
- `shred_paper` - Put the paper into the shredder.
|
| 37 |
+
- `item_classification` - Place the stationery in the yellow box and the electronics in the blue box.
|
| 38 |
+
- `stack_bowls` - Put the blue bowl into the beige bowl, and put the green bowl into the blue bowl.
|
| 39 |
+
|
| 40 |
+
### Embodiments
|
| 41 |
+
|
| 42 |
+
- **ARX5** - Single-arm with triple camera setup (wrist + global + right-side views)
|
| 43 |
+
- **UR5** - Single-arm with dual camera setup (wrist + global views)
|
| 44 |
+
- **ALOHA** - Dual-arm with triple wrist camera setup (left wrist + right wrist + global views)
|
| 45 |
+
- **DOS-W1** - Dual-arm with triple wrist camera setup (left wrist + right wrist + global views)
|
| 46 |
+
|
| 47 |
+
## Dataset Structure
|
| 48 |
+
|
| 49 |
+
### Hierarchy
|
| 50 |
+
|
| 51 |
+
The dataset is organized by tasks, with each task containing multiple demonstration episodes:
|
| 52 |
+
|
| 53 |
+
```
|
| 54 |
+
.
|
| 55 |
+
├── <task_name>/ # e.g., arrange_the_flowers, fold_t_shirt
|
| 56 |
+
│ ├── task_desc.json # Task description
|
| 57 |
+
│ ├── meta/ # Task-level metadata
|
| 58 |
+
│ │ ├── task_info.json
|
| 59 |
+
│ └── data/ # Episode data
|
| 60 |
+
│ ├── episode_000000/ # Individual episode
|
| 61 |
+
│ │ ├── meta/
|
| 62 |
+
│ │ │ └── episode_meta.json # Episode metadata
|
| 63 |
+
│ │ ├── states/
|
| 64 |
+
│ │ │ # for single-arm (ARX5, UR5)
|
| 65 |
+
│ │ │ ├── states.jsonl # Single-arm robot states
|
| 66 |
+
│ │ │ # for dual-arm (ALOHA, DOS-W1)
|
| 67 |
+
│ │ │ ├── left_states.jsonl # Left arm states
|
| 68 |
+
│ │ │ └── right_states.jsonl # Right arm states
|
| 69 |
+
│ │ └── videos/
|
| 70 |
+
│ │ # Video configurations varies by robot model:
|
| 71 |
+
│ │ # ARX5
|
| 72 |
+
│ │ ├── cam_arm_rgb.mp4 # Wrist view
|
| 73 |
+
│ │ ├── cam_global_rgb.mp4 # Global view
|
| 74 |
+
│ │ └── cam_side_rgb.mp4 # Side view
|
| 75 |
+
│ │ # UR5
|
| 76 |
+
│ │ ├── cam_global_rgb.mp4 # Global view
|
| 77 |
+
│ │ └── cam_arm_rgb.mp4 # Wrist view
|
| 78 |
+
│ │ # ALOHA
|
| 79 |
+
│ │ ├── cam_high_rgb.mp4 # Global view
|
| 80 |
+
│ │ ├── cam_left_wrist_rgb.mp4 # Left wrist view
|
| 81 |
+
│ │ └── cam_right_wrist_rgb.mp4 # Right wrist view
|
| 82 |
+
│ │ # DOS-W1
|
| 83 |
+
│ │ ├── cam_high_rgb.mp4 # Global view
|
| 84 |
+
│ │ ├── cam_left_wrist_rgb.mp4 # Left wrist view
|
| 85 |
+
│ │ └── cam_right_wrist_rgb.mp4 # Right wrist view
|
| 86 |
+
│ ├── episode_000001/
|
| 87 |
+
│ └── ...
|
| 88 |
+
├── convert_to_lerobot.py # Conversion script
|
| 89 |
+
└── README.md
|
| 90 |
+
```
|
| 91 |
+
|
| 92 |
+
### Metadata Schema
|
| 93 |
+
|
| 94 |
+
`task_info.json`
|
| 95 |
+
|
| 96 |
+
```json
|
| 97 |
+
{
|
| 98 |
+
"task_desc": {
|
| 99 |
+
"task_name": "arrange_flowers", // Task identifier
|
| 100 |
+
"prompt": "Put the 4 flowers into the vase.",
|
| 101 |
+
"description": "...",
|
| 102 |
+
"scoring": "...", // Scoring criteria
|
| 103 |
+
"task_tag": [ // Task characteristics
|
| 104 |
+
"repeated",
|
| 105 |
+
"single-arm",
|
| 106 |
+
"ARX5",
|
| 107 |
+
"precise3d"
|
| 108 |
+
]
|
| 109 |
+
},
|
| 110 |
+
"video_info": {
|
| 111 |
+
"fps": 30, // Video frame rate
|
| 112 |
+
"ext": "mp4", // Video format
|
| 113 |
+
"encoding": {
|
| 114 |
+
"vcodec": "libx264", // Video codec
|
| 115 |
+
"pix_fmt": "yuv420p" // Pixel format
|
| 116 |
+
}
|
| 117 |
+
}
|
| 118 |
+
}
|
| 119 |
+
```
|
| 120 |
+
|
| 121 |
+
`episode_meta.json`
|
| 122 |
+
|
| 123 |
+
```json
|
| 124 |
+
{
|
| 125 |
+
"start_time": 1750405586.3430033, // Unix timestamp (start)
|
| 126 |
+
"end_time": 1750405642.5247612, // Unix timestamp (end)
|
| 127 |
+
"frames": 1672, // Total video frames
|
| 128 |
+
"robot_id": "rc_arx5_5", // Robot identifier
|
| 129 |
+
"features": {
|
| 130 |
+
"cam_global": { // Camera name info
|
| 131 |
+
"intrinsics": [], // Intrinsics
|
| 132 |
+
"extrinsics": { // Extrinsics
|
| 133 |
+
"arms": {
|
| 134 |
+
"arm": [] // Extrinsic relative to arm
|
| 135 |
+
}
|
| 136 |
+
}
|
| 137 |
+
}
|
| 138 |
+
}
|
| 139 |
+
}
|
| 140 |
+
```
|
| 141 |
+
|
| 142 |
+
### Robot States Schema
|
| 143 |
+
|
| 144 |
+
Each episode contains states data stored in JSONL format. Depending on the embodiment, the structure differs slightly:
|
| 145 |
+
|
| 146 |
+
- **Single-arm robots (ARX5, UR5)** → `states.jsonl`
|
| 147 |
+
- **Dual-arm robots (ALOHA, DOS-W1)** → `left_states.jsonl` and `right_states.jsonl`
|
| 148 |
+
|
| 149 |
+
Each file records the robot’s proprioceptive signals per frame, including joint angles,
|
| 150 |
+
end-effector poses, gripper states, and timestamps. The exact field definitions and coordinate conventions vary by
|
| 151 |
+
platform,
|
| 152 |
+
as summarized below.
|
| 153 |
+
|
| 154 |
+
#### ARX5
|
| 155 |
+
|
| 156 |
+
| Data Name | Data Key | Shape | Semantics |
|
| 157 |
+
|:----------------:|:----------------:|:-----:|:---------------------------------------------------------------------------------------------------------------------------------------------------------------:|
|
| 158 |
+
| Joint control | joint_positions | (6,) | Joint angle (in radians) from the base to the end effector. |
|
| 159 |
+
| Joint velocity | joint_velocities | (6,) | Speed of 6 joint. |
|
| 160 |
+
| Joint effort | efforts | (7,) | Effort of 6 joints and gripper. (Provided by official API, precision not guaranteed. |
|
| 161 |
+
| 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. |
|
| 162 |
+
| Gripper control | gripper_width | (1,) | Actual gripper width measurement in meter. |
|
| 163 |
+
| Gripper velocity | gripper_velocity | (1,) | Speed of gripper. |
|
| 164 |
+
| Time stamp | timestamp | (1,) | Floating point timestamp (in milliseconds) of each frame. |
|
| 165 |
+
|
| 166 |
+
#### UR5
|
| 167 |
+
|
| 168 |
+
| Data Name | Data Key | Shape | Semantics |
|
| 169 |
+
|:---------------:|:---------------:|:-----:|:---------------------------------------------------------------------------------------------------------------------------------------------------------------:|
|
| 170 |
+
| Joint control | joint_positions | (6,) | Joint angle (in radians) from the base to the end effector. |
|
| 171 |
+
| 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. |
|
| 172 |
+
| Gripper control | gripper_width | (1,) | Actual gripper width measurement in meter. |
|
| 173 |
+
| Time stamp | timestamp | (1,) | Floating point timestamp (in milliseconds) of each frame. |
|
| 174 |
+
|
| 175 |
+
#### DOS-W1
|
| 176 |
+
|
| 177 |
+
| Data Name | Data Key | Shape | Semantics |
|
| 178 |
+
|:---------------:|:---------------:|:-----:|:---------------------------------------------------------------------------------------------------------------------------------------------------------------:|
|
| 179 |
+
| Joint control | joint_positions | (6,) | Joint angle (in radians) from the base to the end effector. |
|
| 180 |
+
| 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. |
|
| 181 |
+
| Gripper control | gripper_width | (1,) | Actual gripper width measurement in meter. |
|
| 182 |
+
| Time stamp | timestamp | (1,) | Floating point timestamp (in milliseconds) of each frame. |
|
| 183 |
+
|
| 184 |
+
#### ALOHA
|
| 185 |
+
|
| 186 |
+
| Data Name | Data Key | Shape | Semantics |
|
| 187 |
+
|:--------------------:|:----------------:|:-----:|:---------------------------------------------------------------------------------------------------------------------------------------------------------------:|
|
| 188 |
+
| Joint control | joint_positions | (6,) | Puppet joint angle (in radians) from the base to the end effector. |
|
| 189 |
+
| Joint velocity | joint_velocities | (7,) | Speed of 6 joint. |
|
| 190 |
+
| Gripper control | gripper_width | (1,) | Actual gripper width measurement in meter. |
|
| 191 |
+
| 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. |
|
| 192 |
+
| Joint effort | efforts | (7,) | Effort of 6 joints and gripper. (Provided by official API, precision not guaranteed. |
|
| 193 |
+
| Master Joint effort | master_efforts | (7,) | Effort of 6 master joints and gripper. (Provided by official API, precision not guaranteed. |
|
| 194 |
+
| Master joint control | master_qpos | (6,) | Master joint angle (in radians) from the base to the end effector. |
|
| 195 |
+
| Time stamp | timestamp | (1,) | Floating point timestamp (in mileseconds) of each frame. |
|
| 196 |
+
|
| 197 |
+
## Convert to LeRobot
|
| 198 |
+
|
| 199 |
+
While you can implement a custom Dataset class to read RoboChallenge data directly, **we strongly recommend converting
|
| 200 |
+
to LeRobot format** to take advantage of [LeRobot](https://github.com/huggingface/lerobot)'s comprehensive data
|
| 201 |
+
processing and loading utilities.
|
| 202 |
+
|
| 203 |
+
The example script **`convert_to_lerobot.py`** converts **ARX5** data to the LeRobot dataset as a example. For other
|
| 204 |
+
robot embodiments (UR5, ALOHA, DOS-W1), you can adapt the script accordingly.
|
| 205 |
+
|
| 206 |
+
### Prerequisites
|
| 207 |
+
|
| 208 |
+
- Python 3.9+ with the following packages:
|
| 209 |
+
- `lerobot==0.1.0`
|
| 210 |
+
- `opencv-python`
|
| 211 |
+
- `numpy`
|
| 212 |
+
- Configure `$LEROBOT_HOME` (defaults to `~/.lerobot` if unset).
|
| 213 |
+
|
| 214 |
+
```bash
|
| 215 |
+
pip install lerobot==0.1.0 opencv-python numpy
|
| 216 |
+
export LEROBOT_HOME="/path/to/lerobot_home"
|
| 217 |
+
```
|
| 218 |
+
|
| 219 |
+
### Usage
|
| 220 |
+
|
| 221 |
+
Run the converter from the repository root (or provide an absolute path):
|
| 222 |
+
|
| 223 |
+
```bash
|
| 224 |
+
python convert_to_lerobot.py \
|
| 225 |
+
--repo-name example_repo \
|
| 226 |
+
--raw-dataset /path/to/example_dataset \
|
| 227 |
+
--frame-interval 1
|
| 228 |
+
```
|
| 229 |
+
|
| 230 |
+
### Output
|
| 231 |
+
|
| 232 |
+
- Frames and metadata are saved to `$LEROBOT_HOME/<repo-name>`.
|
| 233 |
+
- At the end, the script calls `dataset.consolidate(run_compute_stats=False)`. If you require aggregated statistics, run
|
| 234 |
+
it with `run_compute_stats=True` or execute a separate stats job.
|