MotionDecode: 1000+ Hours of Humanoid Robot Motion Capture Data

1000+ hours of high-precision optical motion capture data recorded at 120 Hz for humanoid robots and dexterous hands. The ChingMu dataset, created by CMRobot, includes data from 15+ real-world scenes and over 500 standardized tasks, with 200+ tracked objects. It provides skeleton, finger, object 6D pose, video, and label modalities in formats like BVH and NPZ.

Use Cases

• Training robot motion imitation models based on the skeleton and finger motion data.
• Developing object manipulation policies for dexterous hands using the 6D object pose tracking.
• Creating realistic humanoid animations for virtual production based on the retargeted motion capture.
• Benchmarking embodied AI agents on the 500+ standardized tasks across different scenes.

Strengths

• Contains over 1000 hours of motion data captured at a high 120 Hz frequency.
• Covers 15+ distinct real-world scenarios and includes 200+ tracked props with 6D poses.
• Provides multiple data modalities including skeleton, finger, object pose, video, and labels.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

CMRobot

Collection Method

High-precision optical motion capture.

Freshness

Last updated 2026-07-06 06:28:44; freshness should be verified.