AlpaSim

AlpaSim is an open-source autonomous vehicle simulation platform from NVIDIA research (NVlabs) for developing and testing end-to-end AV policies. It simulates sensor data, vehicle dynamics, and traffic scenarios in a closed loop. The latest listed release is v2026.5, and the code is licensed under Apache 2.0.

What It Is

AlpaSim is a Python-based research simulator that lets users run driving policies in a closed-loop testbed. Documented use cases include algorithm validation, safety analysis of edge cases, performance benchmarking and regression testing, and debugging complex driving behavior. Components communicate through a modular gRPC interface, so researchers can swap in custom implementations.

Rendering and Sensor Fidelity

The renderer is a pluggable service. NuRec is supported by default, and OmniDreams video-model rendering is available through FlashDreams. Camera feeds have configurable field of view, resolution, and frame rate, and the simulator models sensor noise and environmental conditions. Stateful video-model rendering is aimed at better fidelity for dynamic objects and non-rigid visuals.

Supported Driving Policies

The README lists Alpamayo-R1, Alpamayo 1.5 checkpoint support, VaVAM, and a provisional Transfuser (LTFv6) integration. A plugin system supports extending AlpaSim with custom models, configs, and tools.

Deployment and Setup

A microservices architecture allows individual components to be scaled and supports multi-node deployments. Local runs use Docker Compose on a single machine by following the tutorial, and cluster deployment uses SLURM tooling in the repository. Documentation covers onboarding, operations, Prometheus telemetry, the ASL log data pipeline, and manual keyboard control of the ego vehicle. Sample scenes are available through a Hugging Face NuRec dataset.