decisionforce/pgdrive

Implements realistic physics simulation and multi-sensor observation spaces while achieving 500+ simulation steps per second, enabling efficient parallel training. Exposes a OpenAI Gym-compatible API with procedurally-generated road networks parameterized by random seeds, supporting customizable traffic density, vehicle dynamics, and environmental conditions across predefined environment sets for benchmarking generalization.

132 stars. No commits in the last 6 months. Available on PyPI.