CtRL-Sim: Reactive and Controllable Driving Agents with Offline Reinforcement Learning

arXiv:2403.19918v1 Announce Type: cross
Abstract: Evaluating autonomous vehicle stacks (AVs) in simulation typically involves replaying driving logs from real-world recorded traffic. However, agents replayed from offline data do not react to the actions of the AV, and their behaviour cannot be easily controlled to simulate counterfactual scenarios. Existing approaches have attempted to address these shortcomings by proposing methods that rely on heuristics or learned generative models of real-world data but these approaches either lack realism or necessitate costly iterative sampling …

abstract agents arxiv autonomous autonomous vehicle avs counterfactual cs.ai cs.lg cs.ro data driving however logs offline react reinforcement reinforcement learning sim simulation stacks traffic type world