
This paper provides autonomous driving datasets and benchmarks for offline reinforcement learning research, incorporating both real-world human driver datasets and synthetic datasets generated by online reinforcement learning agents. The authors also propose a unified partially observable Markov decision process (POMDP) that can be applied across various driving scenarios.


coremsg

ad4rl-autonomous-driving-benchmarks-for-offline-reinforcement-learning-with-real-world-and-synthetic-datasets


AD4RL: Autonomous Driving Benchmarks for Offline Reinforcement Learning with Real-World and Synthetic Datasets
