Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
size_categories:
|
3 |
+
- 1K<n<10K
|
4 |
+
---
|
5 |
+
## Dataset Description
|
6 |
+
|
7 |
+
This is a fully annotated, synthetically generated dataset consisting of 1,000 demonstrations of a single Franka Panda robot arm performing a fixed-order three-cube stacking task in Isaac Lab. The robot consistently stacks cubes in the order: blue (bottom) → red (middle) → green (top).
|
8 |
+
|
9 |
+
The dataset was produced using the following pipeline:
|
10 |
+
- Collected 10 human teleoperation demonstrations of the stacking task.
|
11 |
+
- Used Isaac Lab’s **Mimic** tool to simulate 1,000 high-quality trajectories in Isaac Sim.
|
12 |
+
- Applied **Cosmos Transfer1** model to augment the RGB visuals from the table camera with photorealistic domain adaptation.
|
13 |
+
|
14 |
+
Each demonstration includes synchronized multimodal data:
|
15 |
+
- RGB videos from both a table-mounted and wrist-mounted camera.
|
16 |
+
- Depth, segmentation, and surface normal maps from the table camera.
|
17 |
+
- Full low-level robot and object states (joints, end-effector, gripper, cube poses).
|
18 |
+
- Action sequences executed by the robot.
|
19 |
+
|
20 |
+
This dataset is ideal for behavior cloning, policy learning, and generalist robotic manipulation research.
|
21 |
+
|
22 |
+
## Intended Usage
|
23 |
+
|
24 |
+
This dataset is intended for:
|
25 |
+
- Training robot manipulation policies using behavior cloning.
|
26 |
+
- Research in generalist robotics and task-conditioned agents.
|
27 |
+
- Sim-to-real transfer studies and visual domain adaptation.
|
28 |
+
|
29 |
+
## Dataset Characterization
|
30 |
+
|
31 |
+
**Data Collection Method**
|
32 |
+
* Human Demonstration (seed data)
|
33 |
+
* Synthetic Simulation (Isaac Lab Mimic)
|
34 |
+
* Visual Augmentation (Cosmos Transfer1)
|
35 |
+
|
36 |
+
10 human teleoperated demonstrations were used to bootstrap a Mimic-based simulation in Isaac Sim. All 1,000 demos are generated automatically followed by domain-randomized visual augmentation.
|
37 |
+
|
38 |
+
## Dataset Format
|
39 |
+
|
40 |
+
Each demo consists of a time-indexed sequence of the following modalities:
|
41 |
+
|
42 |
+
### Actions
|
43 |
+
- 7D vector: 6D relative end-effector motion + 1D gripper action
|
44 |
+
|
45 |
+
### Observations
|
46 |
+
- `robot_states`: Joint positions, velocities, and gripper open/close state
|
47 |
+
- `ee_states`: End-effector 6-DOF pose
|
48 |
+
- `cube_states`: 6-DOF poses (position + quaternion) for blue, red, and green cubes
|
49 |
+
- `images.table_camera`:
|
50 |
+
- 200×200 RGB (visually augmented using Cosmos Transfer1)
|
51 |
+
- 200×200 Depth
|
52 |
+
- 200×200 Segmentation mask
|
53 |
+
- 200×200 Surface normal map
|
54 |
+
- `images.wrist_camera`:
|
55 |
+
- 200×200 RGB
|