arXiv Robotics — research abstracts· Yuqi Ye, Shangkun Sun, Junhong Lin, Jiayi Zhao, Changhao Peng, Wei Zheng, Guoqing Liu, Tiesong Zhao, Wei Gao·· 1 days agoEditorial score66
Sometimes You Gotta Run Before You Can Walk: Run-then-Walk Scheduling Strategy for VLM Autonomous Driving
Sometimes You Gotta Run Before You Can Walk: Run-then-Walk Scheduling Strategy for VLM Autonomous Driving
Summary
This research proposes a two-stage reward scheduling strategy called 'Run-then-Walk' for VLM-based autonomous driving planners. It addresses the limitations of existing GRPO-style reinforcement learning methods by separating progress discovery and safety repair. The approach achieves better performance and faster convergence, validated on multiple benchmarks with reduced training epochs.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
Source：arXiv Robotics — research abstracts · arxiv.org