Physicalai Robotics Locomanipulation Grail

A dataset of physics-validated 4D human-object interaction trajectories for the Unitree G1 humanoid robot, generated by the GRAIL pipeline. It was created by NVIDIA and last updated on HuggingFace in June 2026. The data includes scenarios such as tabletop pickup, ground manipulation, and navigating stairs and slopes.

Use Cases

• Training reinforcement learning agents for humanoid robot manipulation based on the described physics-validated trajectories.
• Benchmarking robot control policies for tasks like tabletop and ground pickup mentioned in the description.
• Developing sim-to-real transfer methods using the 4D interaction data generated from 3D assets and video priors.
• Studying humanoid locomotion in complex environments like stairs and slopes as listed in the dataset overview.

Strengths

• Data is described as 'physics-validated', suggesting a level of physical realism.
• Covers multiple complex scenarios for a humanoid robot, including sitting, curb navigation, slope, and stairs.
• Generated by an end-to-end pipeline (GRAIL) from NVIDIA, a known entity in AI and robotics.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file size are unknown, which may limit suitability assessment.
• The full description is truncated, requiring a visit to an external page for complete details.

Provenance

Source

NVIDIA

Collection Method

Generated by the GRAIL (Generating Humanoid Loco-Manipulation from 3D Assets and Video Priors) pipeline.

Freshness

Last updated 2026-06-04 04:43:17; freshness should be verified.