MetaDrive: Composing Diverse Driving Scenarios for Generalizable Reinforcement Learning

Quanyi Li1,4*,  Zhenghao Peng1*,  Lan Feng3Qihang Zhang1Zhenghai Xue1Bolei Zhou2 
1The Chinese University of Hong Kong, 2University of California, Los Angeles 3ETH Zurich
4Centre for Perceptual and Interactive Intelligence
IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022
Webpage | Code | Paper


MetaDrive Simulator

To facilitate the research of generalizable reinforcement learning, we develop an open-source, highly efficient and flexible driving simulator MetaDrive, which holds the following key features:

We construct a variety of RL tasks and baselines in both single-agent and multi-agent settings, including benchmarking generalizability across unseen scenes, safe exploration, and learning multi-agent traffic.

Please refer to the technical report or the code repo for more information.
  title={Metadrive: Composing diverse driving scenarios for generalizable reinforcement learning},
  author={Li, Quanyi and Peng, Zhenghao and Feng, Lan and Zhang, Qihang and Xue, Zhenghai and Zhou, Bolei},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},