Virtual to Real Reinforcement Learning for Autonomous Driving

TL;DR

This paper introduces a realistic translation network that enables reinforcement learning policies trained in virtual environments to effectively transfer and operate in real-world autonomous driving scenarios.

Contribution

A novel virtual-to-real translation network is proposed, facilitating the transfer of reinforcement learning-based driving policies from simulation to real-world environments.

Findings

The VR RL approach successfully adapts to real-world driving data.

First known case of RL-trained driving policy transferring effectively to real-world driving.

The translation network maintains scene structure during virtual-to-real conversion.

Abstract

Reinforcement learning is considered as a promising direction for driving policy learning. However, training autonomous driving vehicle with reinforcement learning in real environment involves non-affordable trial-and-error. It is more desirable to first train in a virtual environment and then transfer to the real environment. In this paper, we propose a novel realistic translation network to make model trained in virtual environment be workable in real world. The proposed network can convert non-realistic virtual image input into a realistic one with similar scene structure. Given realistic frames as input, driving policy trained by reinforcement learning can nicely adapt to real world driving. Experiments show that our proposed virtual to real (VR) reinforcement learning (RL) works pretty well. To our knowledge, this is the first successful case of driving policy trained by reinforcement learning that can adapt to real world driving data.