metadriverse/metadrive

Health v2 · maintenance only

Adoption not part of the score

1235 stars · 197 forks observed · 2026-08-28

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

MetaDrive is an open-source, lightweight driving simulator built for AI and autonomy research, supporting compositional scene synthesis and reconstruction of real-world driving logs. It provides realistic physics and multiple sensor modalities (RGB, depth, semantic, LiDAR point cloud) and runs at over 1000 FPS on standard PCs.

Use cases

• train reinforcement learning agents for autonomous driving
• benchmark generalizable RL across unseen driving scenes
• simulate safe exploration and safety-critical driving scenarios
• reconstruct driving environments from Waymo or nuPlan datasets
• generate synthetic driving scenes with custom road maps and traffic
• collect camera and LiDAR sensor data for perception research
• run multi-agent driving behavior simulations

When to choose

• you need a fast, easy-to-install driving simulator for RL research
• you want to train or evaluate driving policies in Python with Gym-style APIs
• you need sensor simulation (RGB, depth, LiDAR) without heavy commercial simulators
• you want to replay real-world datasets like Waymo Open Dataset in simulation
• you study safe RL or multi-agent driving behaviors

When to avoid

• you need photorealistic rendering comparable to CARLA or commercial simulators
• you require full vehicle dynamics validation for hardware deployment
• you need non-driving robotics simulation
• you want a plug-and-play product rather than a research codebase

Facets

library · maturity active

simulation machine-learning reinforcement-learning computer-vision game-engine autonomous-vehicles simulation reinforcement-learning machine-learning robotics windows python cross-platform driving-simulator autonomous-driving reinforcement-learning-environment gym-environment scene-generation sensor-simulation safe-rl multi-agent linux macos

Member repositories

For agents

markdown · JSON · MCP: product_card(name="metadriverse/metadrive")