Predicting the future for autonomous driving

Anthony Hu

University of Cambridge/Wayve

Autonomous driving has been a long-standing challenge in artificial intelligence (AI). Fundamentally different from recent successes in computer vision (ImageNet), natural language processing (GTP-3), or strategy games (AlphaZero, OpenAI Five), autonomous vehicles are Embodied AIs where the software is equipped with a physical body that interacts with the real-world. I will present what are the key differences of such a system, our approach to autonomous driving at Wayve, as well as our recent findings in learning a predictive world-model to anticipate the future for safer driving.