autonomousvision/navsim

Health v2 · maintenance only

Adoption not part of the score

1076 stars · 128 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

NAVSIM is a data-driven pseudo-simulation framework and benchmark for autonomous vehicle planning, evaluating driving agents non-reactively on logged sensor data. It powers public leaderboards (navtest/navhard) for end-to-end driving research from the University of Tübingen, NVIDIA Research, and collaborators.

Use cases

• benchmark end-to-end autonomous driving planners
• evaluate driving policies without full reactive simulation
• run pseudo-simulation on logged driving datasets
• submit entries to the NAVSIM challenge leaderboard
• train and validate self-driving planning models
• compare planning agents on standardized metrics

When to choose

• you research autonomous driving planning and need a standardized benchmark
• you want leaderboard-comparable evaluation without building a full simulator
• you work with nuPlan/OpenScene-style logged sensor data

When to avoid

• you need closed-loop reactive simulation with interactive traffic
• you need a production-grade driving simulator for game or robotics deployment
• you don't work with large autonomous driving datasets

Facets

library · maturity active

simulation machine-learning benchmarking computer-vision autonomous-vehicles machine-learning simulation computer-vision python autonomous-driving planning self-driving pseudo-simulation benchmark navsim dataset linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="autonomousvision/navsim")