PhysicalAI NuRec: NVIDIA Robotics Assets with USD and Occupancy Maps

NVIDIA released the Physical AI NuRec dataset in March 2026 to facilitate end-to-end robotic model training. It provides 3D assets in Universal Scene Description (USD) format, including mesh components for collision detection and occupancy maps for navigation within the Isaac Sim environment.

Use Cases

• Training object detection models using 3DGUT components for realistic visual rendering
• Developing collision avoidance systems using the provided mesh components for physics simulation
• Generating navigation paths for Autonomous Mobile Robots (AMR) using occupancy maps and MobilityGen

Strengths

• Native USD format for seamless integration with NVIDIA Isaac Sim
• Dual-layer assets providing both mesh for collisions and 3DGUT for rendering
• Licensed under CC BY 4.0 for open research and commercial use

Limitations

• Ecosystem lock-in to NVIDIA Isaac Sim for full feature utilization
• Lack of tabular metadata or explicit class labels in the primary description

Provenance

Source

NVIDIA

Collection Method

Synthetic asset generation for robotics simulation

Freshness

Last updated March 2026.

Geography

United States