Artificial Driving Intelligence: Are we over-engineering self driving cars?

Abstract: Billions of dollars and over a decade of engineering resources have been poured into solving the problem of autonomous driving. So why do we not have self-driving cars on the road in a meaningful way today? Over the last 5-10 years, the advances in computer vision, machine learning and reinforcement learning have blown us all away e.g. super-human level performance in ImageNet classification problems and Go, using end-to-end deep learning. Perhaps it is these very methods which the autonomous driving community must turn to.

In this talk, I wish to describe Wayve's philosophy of bridging the gap between traditional robotic approaches and cutting edge machine learning. I will demonstrate some of our technology around robotic learning on an actual full sized road vehicle, including deep reinforcement learning and transfer learning.

Bio: Amar is co-founder and CEO at Wayve Technologies, a UK based AI first robotics startup. Prior to starting Wayve, Amar completed a PhD in machine learning with Zoubin Ghahramani in Cambridge, and a postdoc with Yoshua Bengio. During his PhD, Amar applied deep machine learning to robotic problems at NASA, for which he was awarded a Planetary Defense Award. His PhD work won him the Qualcomm Innovation Fellowship and a EPSRC Doctoral Prize. Prior to his PhD, Amar was a quantitative strategist at Goldman Sachs.