This thesis presents two learning based approaches to solve the autonomous driving problem: end-to-end imitation learning and direct visual perception. Imitation learning uses expert demonstrations to build a policy that ...