Research papers · 2022.03.11
Automatically Learning Fallback Strategies with Model-Free Reinforcement Learning in Safety-Critical Driving Scenarios
Automatically Learning Fallback Strategies with Model-Free Reinforcement Learning in Safety-Critical Driving Scenarios
Overview
An additional reward based on distances between agent trajectories encourages exploration beyond a dominant policy. An autonomous-driving example demonstrates learning alternative behavioral modes that can serve as fallback strategy candidates.
Author-supplied source information
Author-supplied information on arXiv.
To appear in proceedings of International Conference on Machine Learning Technologies (ICMLT) 2022
Version: arXiv v1 — manuscript revision