Learning a Driving Simulator

TL;DR

This paper presents a novel approach to driving simulation using variational autoencoders, GANs, and RNNs to generate realistic road video sequences for self-driving car research.

Contribution

It introduces a method combining variational autoencoders, GANs, and RNNs to learn and simulate road scenes without relying on pixel-space cost functions.

Findings

Generated realistic video sequences over multiple frames

Effective embedding of road frames using GANs and VAEs

Transition model maintains visual fidelity without pixel-space optimization

Abstract

Comma.ai's approach to Artificial Intelligence for self-driving cars is based on an agent that learns to clone driver behaviors and plans maneuvers by simulating future events in the road. This paper illustrates one of our research approaches for driving simulation. One where we learn to simulate. Here we investigate variational autoencoders with classical and learned cost functions using generative adversarial networks for embedding road frames. Afterwards, we learn a transition model in the embedded space using action conditioned Recurrent Neural Networks. We show that our approach can keep predicting realistic looking video for several frames despite the transition model being optimized without a cost function in the pixel space.