Abstract

Abstract

End-to-end machine learning has enabled many breakthroughs in computer vision over the last couple of years. At Wayve, it has enabled us to learn driving policies for autonomous vehicles without relying on HD-maps or hand-coded rules. In this talk I will present a self driving software stack for learning meaningful state representations with computer vision (semantics, depth and motion) and driving policies using conditional imitation learning from expert drivers. These models can drive on public urban roads, never seen before during training after learning from only 30 hours of real-world demonstrations.

Bio

Corina Gurau is a researcher at Wayve applying machine learning and computer vision to self-driving vehicles technology. She finished her DPhil at the Oxford Robotics Institute, where she worked on perception for robots operating outdoors, in diverse environmental conditions. Her thesis focused on understanding and remedying failure modes of ML-based vision systems. Before that she finished a BSc in Computer Science at Jacobs University Bremen in Germany.