robotics-diffusion-transformer
commited on
Commit
•
74d1168
1
Parent(s):
87cc448
Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,48 @@
|
|
1 |
-
---
|
2 |
-
license: mit
|
3 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: mit
|
3 |
+
---
|
4 |
+
|
5 |
+
# Dataset Card
|
6 |
+
|
7 |
+
This is the fine-tuning dataset used in the paper *RDT-1B: a Diffusion Foundation Model for Bimanual Manipulation*.
|
8 |
+
|
9 |
+
## Source
|
10 |
+
|
11 |
+
- Project Page: https://rdt-robotics.github.io/rdt-robotics/
|
12 |
+
- Paper:
|
13 |
+
- Code:
|
14 |
+
- Model: https://huggingface.co/robotics-diffusion-transformer/rdt-1b
|
15 |
+
|
16 |
+
## Getting Started
|
17 |
+
|
18 |
+
Download all archive files and use the following command to extract:
|
19 |
+
```bash
|
20 |
+
cat rdt_data.tar.gz.* | tar -xzvf -
|
21 |
+
```
|
22 |
+
|
23 |
+
In the resulting `rdt_data` folder, each sub-folder is a task. In each task sub-folder, we have HDF5 files and the instruction JSON file, as illustrated below:
|
24 |
+
```
|
25 |
+
rdt_data/---task_1/---episode_1.hdf5
|
26 |
+
| |
|
27 |
+
|-task_2/ |-episode_2.hdf5
|
28 |
+
| |
|
29 |
+
|-task_3/ |-...
|
30 |
+
| |
|
31 |
+
|-... |-expanded_instruction_gpt-4-turbo.json
|
32 |
+
```
|
33 |
+
Each HDF5 file corresponds to a trajectory/episode of the task, which has the following keys:
|
34 |
+
- `observations`:
|
35 |
+
- `qpos`: joint positions of the two robot arms, (TRAJ_LEN, 14); the seventh and fourteenth joints are gripper joint angles
|
36 |
+
- `images`:
|
37 |
+
- `cam_high`: RGB images from the exterior camera at each time step, (TRAJ_LEN, 480, 640, 3)
|
38 |
+
- `cam_left_wrist`: RGB images from the left-wrist camera at each time step, (TRAJ_LEN, 480, 640, 3)
|
39 |
+
- `cam_right_wrist`: RGB images from the right-wrist camera at each time step, (TRAJ_LEN, 480, 640, 3)
|
40 |
+
- `action`: desired joint positions of the two robot arms at the next time step,
|
41 |
+
(TRAJ_LEN, 14); Note that this is slightly different from the actual joint positions at the next time step
|
42 |
+
|
43 |
+
Note: The number of HDF5 episode files is not necessarily consecutive. `TRAJ_LEN` may vary from episode to episode.
|
44 |
+
|
45 |
+
Each JSON file corresponds to the annotated language instructions of the task, which has the following keys:
|
46 |
+
- `instruction`: original human-annotated instructions; its value is a string
|
47 |
+
- `expanded_instruction`: instructions generated by GPT-4-Turbo by expanding the original one; its value is a list of strings
|
48 |
+
- `simplified_instruction`: instructions generated by GPT-4-Turbo by simplifying the original one; its value is a list of strings
|