omniretarget commited on
Commit
be52ebc
·
1 Parent(s): bcf486c

Add README

Browse files
Files changed (2) hide show
  1. README.md +78 -0
  2. assets/teaser.mp4 +3 -0
README.md ADDED
@@ -0,0 +1,78 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ language: en
4
+ tags:
5
+ - robotics
6
+ - motion-retargeting
7
+ - reinforcement-learning
8
+ - humanoid
9
+ - trajectory-optimization
10
+ ---
11
+
12
+ # OmniRetarget Dataset: Humanoid Loco-Manipulation & Scene Interaction
13
+
14
+ This dataset contains motion trajectories of a G1 humanoid robot interacting with objects and complex terrains. It was generated by **OMNIRETARGET**, an interaction-preserving data generation engine that produces high-quality, kinematically feasible trajectories free of common artifacts like foot-skating and penetration.
15
+
16
+ <div align="center">
17
+ <video autoplay loop muted controls width="70%">
18
+ <source src="https://huggingface.co/datasets/omniretarget/OmniRetarget_Dataset/resolve/main/assets/teaser.mp4" type="video/mp4">
19
+ </video>
20
+ </div>
21
+
22
+ ## Dataset Structure
23
+
24
+ | Subset | Description | Source Data | Duration (hours) |
25
+ | ------------------------ | --------------------------------------------------- | --------------- | ---------------- |
26
+ | `robot-object/` | Motions of the robot carrying objects. | OMOMO | 2.78 |
27
+ | `robot-terrain/` | Dynamic motions of the robot climbing challenging terrains. | In-house MoCap | 0.5 |
28
+ | `robot-object-terrain/` | Motions involving both object and terrain interaction. | In-house MoCap | 0.5 |
29
+ | **Total** | | | **3.78** |
30
+
31
+ Additionally, the `models/` directory contains all the necessary URDF, SDF, and OBJ assets for visualization. These are not required for loading or training with the trajectory data.
32
+
33
+ ## Data Format
34
+
35
+ Each `.npz` file contains a single trajectory with two keys:
36
+ - **`fps`**: Frames per second.
37
+ - **`qpos`**: A NumPy array of shape `[T, D]` representing the system state over `T` timesteps. The vector is structured as follows:
38
+ - **Robot Pose (36D):**
39
+ - Floating Base `[qw, qx, qy, qz, x, y, z]` (7D)
40
+ - Joint Positions (29D)
41
+ - **Object Pose (7D, optional):**
42
+ - `[qw, qx, qy, qz, x, y, z]`
43
+ - The total dimension `D` is 36 for motions without an object, and 43 with an object.
44
+
45
+ ## Quick Usage
46
+ ```bash
47
+ # Clone the repository
48
+ git lfs install
49
+ git clone https://huggingface.co/datasets/omniretarget/OmniRetarget_Dataset
50
+
51
+ # Load data
52
+ pip install numpy
53
+ import glob, numpy as np
54
+ paths = glob.glob("robot-object/*.npz")
55
+ with np.load(paths[0]) as data:
56
+ qpos = data["qpos"] # (T, D)
57
+ fps = float(data["fps"]) # e.g., 30.0
58
+ ```
59
+
60
+ ## Visualize (optional)
61
+ A `visualize.py` script using Drake and Meshcat is provided.
62
+ ```bash
63
+ # Install dependencies
64
+ pip install drake
65
+
66
+ # Set `task` inside the script: "object" | "terrain" | "object-terrain"
67
+ python visualize.py
68
+ ```
69
+
70
+ ## Citation
71
+ ```bibtex
72
+ @inproceedings{Yang2026OmniRetarget,
73
+ title={OmniRetarget: Interaction-Preserving Data Generation for Humanoid Whole-Body Loco-Manipulation and Scene Interaction},
74
+ author={Yang, Lujie and Huang, Xiaoyu and Wu, Zhen and Kanazawa, Angjoo and Abbeel, Pieter and Sferrazza, Carmelo and Liu, C. Karen and Duan, Rocky and Shi, Guanya},
75
+ booktitle={arXiv},
76
+ year={2026}
77
+ }
78
+ ```
assets/teaser.mp4 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:866b14534f597b0e2e3230518523b90f4069d2b6baefcf2e4998e6fa057fdf79
3
+ size 2642654