lightcone02/OmniContact-Dataset

OmniContact Dataset: Contact-Rich Humanoid Object Interaction

Project Page

This dataset contains human-object interaction motion capture data and processed G1 humanoid trajectories for contact-rich box manipulation and soccer-style interactions. The main entry point is the provided visualizer, which lets users inspect the processed NPZ trajectory, original BVH motion, object pose, and contact labels side by side.

Carry box

_{BVH source}

_{G1 retarget}

Push box

_{BVH source}

_{G1 retarget}

Slide box

_{BVH source}

_{G1 retarget}

Kick ball

_{BVH source}

_{G1 retarget}

Relocate ball

_{BVH source}

_{G1 retarget}

Dataset Structure

Subset	Description	Source Data	Processed Trajectories
npz/box/case1_carry/	G1 trajectories for carrying boxes.	In-house MoCap	217
npz/box/case2_push/	G1 trajectories for pushing boxes with hand contact.	In-house MoCap	165
npz/box/case3_slide/	G1 trajectories for sliding boxes.	In-house MoCap	118
npz/soccer/case1_kick_forward/	Forward soccer-ball kicking motions.	In-house MoCap	30
npz/soccer/case2_kick_left/	Left soccer-ball kicking motions.	In-house MoCap	30
npz/soccer/case3_kick_right/	Right soccer-ball kicking motions.	In-house MoCap	30
npz/soccer/case4_carry/	Carrying the soccer ball.	In-house MoCap	51
npz/soccer/case5_pickball_kick/	Picking up and kicking the soccer ball.	In-house MoCap	59
Total			700

Additional directories:

• raw_mocap/box/ and raw_mocap/soccer/ contain sanitized source captures with motion_actor.bvh, motion_actor.csv, object-pose CSV files (prop_*.csv or object_pose_*.csv), and capture_meta.json.
• assets/ contains the G1 URDF and object meshes used by the visualizer.
• bvh_cache/ contains cached BVH parses for faster loading.
• metadata/raw_mocap_manifest.csv maps capture IDs to cases and raw files.
• metadata/splits.csv provides the recommended 70/15/15 train/val/test split.
• demo/mocap_task_videos/ contains one rendered mocap preview video for each task.
• demo/g1_videos/ contains recorded G1 trajectory previews for matched examples.
• demo/g1_glbs/ contains serialized G1 preview data exported from processed .npz trajectories (currently JSON).

Data Format

Each processed .npz file contains one trajectory. Common keys include:

• fps: Frames per second.
• base_pos_w: Floating-base position, shape [T, 3].
• base_quat_w: Floating-base orientation in wxyz order, shape [T, 4].
• joint_pos: G1 joint positions, shape [T, 29].
• body_pos_w: Body positions, shape [T, 39, 3].
• body_quat_w: Body orientations in wxyz order, shape [T, 39, 4].
• object_pos_w: Object position, shape [T, 3].
• object_quat_w: Object orientation in wxyz order, shape [T, 4].
• contact_info: Binary contact labels, shape [T, 4, 1].

The contact-label order is:

left_ankle, right_ankle, left_wrist, right_wrist

Raw mocap folders keep only the files needed to reconstruct motion and object pose. Operator metadata, internal configs, logs, validation reports, quality reports, and ZIP files are excluded.

Quick Usage

# After cloning or downloading this dataset
pip install numpy

import numpy as np

path = "npz/box/case1_carry/20260330000166_1_1775007076_with_contact.npz"
with np.load(path, allow_pickle=False) as data:
    joint_pos = data["joint_pos"]          # (T, 29)
    base_pos = data["base_pos_w"]          # (T, 3)
    base_quat = data["base_quat_w"]        # (T, 4), wxyz
    object_pos = data["object_pos_w"]      # (T, 3)
    contact = data["contact_info"]         # (T, 4, 1)
    fps = float(data["fps"][0])

Use metadata/splits.csv to select the recommended train/val/test split by capture_id.

Visualize

The repository provides visualize_npz_bvh_minimal.py, a local browser visualizer based on Viser. It shows the processed G1 trajectory and original BVH skeleton side by side, with object pose and contact highlights.

# Install visualization dependencies
pip install numpy viser yourdfpy

Box example:

python visualize_npz_bvh_minimal.py \
  --data npz/box/case1_carry/20260330000336_1_1775817056_with_contact.npz

Soccer example:

python visualize_npz_bvh_minimal.py \
  --data npz/soccer/case1_kick_forward/20260330000532_1_1775265300_with_contact.npz

Then open http://localhost:8092.

To browse all matched NPZ/BVH samples:

python visualize_npz_bvh_minimal.py

Useful options:

• --max-frames 600: load a short prefix for quick inspection.
• --playback-speed 0.5: slow down playback.
• --sync-mode frame: synchronize NPZ and BVH by frame index.
• --align-roots: align the first BVH root to the first NPZ root.
• --bvh-line-width 1: render a thinner BVH skeleton.

Citation

@misc{yu2026omnicontactchainingmetaskillscontact,
      title={OmniContact: Chaining Meta-Skills via Contact Flow for Generalizable Humanoid Loco-Manipulation}, 
      author={Runyi Yu and Xiaoyi Lin and Ji Ma and Yinhuai Wang and Koukou Luo and Jiahao Ji and Huayi Wang and Wenjia Wang and Runhan Zhang and Ping Tan and Ting Wu and Ruoli Dai and Qifeng Chen and Lei Han},
      year={2026},
      eprint={2606.26201},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2606.26201}, 
}