ChatSTL: A Framework of Translation from Natural Language to Signal Temporal Logic Specifications for Autonomous Vehicle Navigation out of Blocked Scenarios
Yujin Wang, Zhaoyan Huang, Shiying Dong, Hongqing Chu, Xiang Yin, Bingzhao Gao · 2024
This paper addresses the quandary that autonomous vehicles may encounter in blocked scenarios, and proposes a novel framework of translation from natural language to Signal Temporal Logic specifications based on Large Language Model, which helps autonomous vehicles navigate out of blocked scenarios through path planning. We utilize the profound capability of reasoning and planning of GPT-3.5, design a set of predicate functions and formulate the prompts in order to generate accurate Signal Temporal Logic specifications describing driving tasks. The feasibility of the proposed framework is demonstrated via a numerical experiment.