End2End Autonomous Driving: Performance, Measurement & Safety

Vijay Badrinarayanan

Wayve, USA

Abstract

The focus of machine learning is quite definitely turning to solving robotics for practical real world applications. Both the industry and academic communities are converging to end2end learnt policies using large transformer architectures & diverse datasets. In this lecture I will cover (a) the basic elements necessary to understand the problem domain, (b) pros and cons of existing attempts to solve the problem, and (c) the rapid progress that has been made in real world autonomous driving using end2end trained models, state of the art techniques to generate data & measure robot performance in order to accelerate real world deployment.