DriverGym: Democratising Reinforcement Learning for Autonomous Driving

TL;DR

DriverGym is an open-source platform that facilitates the development and validation of reinforcement learning algorithms for autonomous driving using real-world data and extensive evaluation tools.

Contribution

It introduces DriverGym, a novel open-source environment compatible with OpenAI Gym, tailored for RL in autonomous driving, with comprehensive data and evaluation protocols.

Findings

Provides over 1000 hours of expert data for training and validation.

Supports reactive and data-driven agent behaviors.

Includes baseline models for behavior cloning and reinforcement learning.

Abstract

Despite promising progress in reinforcement learning (RL), developing algorithms for autonomous driving (AD) remains challenging: one of the critical issues being the absence of an open-source platform capable of training and effectively validating the RL policies on real-world data. We propose DriverGym, an open-source OpenAI Gym-compatible environment specifically tailored for developing RL algorithms for autonomous driving. DriverGym provides access to more than 1000 hours of expert logged data and also supports reactive and data-driven agent behavior. The performance of an RL policy can be easily validated on real-world data using our extensive and flexible closed-loop evaluation protocol. In this work, we also provide behavior cloning baselines using supervised learning and RL, trained in DriverGym. We make DriverGym code, as well as all the baselines publicly available to further stimulate development from the community.