1,727 hours
#1 of 13
of driving data in Nvidia's PhysicalAI-AV dataset (2025), the largest of the 13 tracked; nuPlan (2024) has 1,282

What the chart shows

The amount of driving data in public autonomous driving benchmarks has grown. Early sets such as KITTI (2012, 6 hours) and nuScenes (2019, 5.6 hours) held under 10 hours; the Waymo Open Dataset (2019) brought 574 hours, nuPlan (2024) 1,282 and Nvidia’s PhysicalAI-AV (2025) 1,727.

Ten of the 13 still hold under 120 hours, including recent end-to-end and reasoning sets. The AI Index cautions that hours do not capture quality or content: simulated driving is not the same as data from real roads.

Key numbers

MeasureValue
Nvidia PhysicalAI-AV (2025)1,727 h
nuPlan (2024)1,282 h
Waymo Open Dataset (2019)574 h
NuPlanQA* (2025)119.44 h
KITTI Vision Benchmark Suite (2012)6 h
DriveLMM-o1* (2025)0.35 h

How to read it

  • Datasets marked * report estimated rather than measured hours.
  • The horizontal axis is logarithmic, so each gridline is ten times the one before.
  • The report text gives about 1,600 hours for PhysicalAI-AV and roughly 500 for the Waymo Open Dataset; the data shows 1,727 and 574, which we use.
Source: AI Index, 2025, via the AI Index Report 2026 (Stanford HAI), figure 2.7.8: Autonomous driving benchmarks/datasets: hours of driving data, 2012–25. Chart and text by AI Adoption.
Numbers checked against the published data on 20 October 2026.
Download the data (CSV) · Cite this chart