Found-RL: foundation model-enhanced reinforcement learning for autonomous driving

Researchers introduced Found-RL, a new platform that dramatically accelerates reinforcement learning for self-driving cars by integrating foundation models. Their key innovation is an asynchronous batch inference framework that decouples heavy VLM reasoning from the simulation loop, solving latency bottlenecks. The system enables a lightweight RL model to achieve near-VLM performance while sustaining real-time inference at approximately 500 FPS—massively faster than traditional approaches that struggle with VLM latency.