Inside the Lab: Training Self-Driving Cars with Reinforcement Learning

This masterclass offers an inside look at how reinforcement learning (RL) is being used to design and train self-driving cars entirely within simulated environments. By relying on high-quality synthetic data instead of costly and time-consuming real-world driving data, this approach presents a scalable and practical pathway for developing autonomous vehicle systems. Participants will be guided through the end-to-end process of building reinforcement learning training pipelines, with a deep dive into the challenges of simulating realistic driving environments and generating diverse training data at scale. The session will highlight how billions of synthetic driving scenarios—including rare and high-risk edge cases—can be efficiently created to improve model robustness and safety. The masterclass will also explore how models trained in simulation can successfully transfer to real-world driving conditions, addressing the critical gap between controlled virtual environments and the complexity of real roads. Attendees will gain practical insights into the tools, techniques, and research strategies shaping the future of autonomous driving.