PhysicalAI-Autonomous-Vehicles: 1,700 Hours of Multi-Sensor Driving Data

NVIDIA's PhysicalAI-Autonomous-Vehicles dataset contains 1,700 hours of multi-sensor driving data for end-to-end autonomous vehicle research. Collected via automatic sensor methods, it focuses on geographic diversity to support Physical AI development as of early 2026.

Use Cases

• Training end-to-end driving models using multi-sensor input streams
• Developing Physical AI systems for real-world navigation
• Benchmarking sensor fusion algorithms across 1,700 hours of footage

Strengths

• 1,700 hours of driving data
• Geographically diverse collection sites
• Automatic sensor-based labeling

Limitations

• Lack of manual annotation verification due to automatic labeling method
• Undefined sensor specifications in metadata

Provenance

Source

NVIDIA

Collection Method

Automatic sensor-based capture

Freshness

Last updated March 2026.

Geography

United States