arxivcs.ROcs.CV2026-07-03
CLEAR: Closed-Loop Reinforcement Learning at Scale for End-to-End Autonomous Driving
Yunxiao Shi, Hong Cai, Mohammad Ghavamzadeh, Fatih Porikli
End-to-end autonomous driving (E2E-AD) aims to directly map raw sensor information to driving actions. Recently, with the rapid advancement of multi-modal large language models (MLLMs), researchers have proposed the paradigm of Vision-Language-Action (VLA) models for E2E-AD, wher…