Acar Celik, Esra (fortiss GmbH Research Institute of the Free State of Bavaria ass), Liu, Xiangzhong (fortiss GmbH Research Institute of the Free State of Bavaria ass), Zhang, Jiajie (fortiss Research Institute of the Free State of Bavaria associat)

Traffic Rule Integration with Temporal Logic in Deep Reinforcement Learning for Behavior Planning

Scheduled for presentation during the Regular Session "Driving based on reinforcement learning" (ThAT6), Thursday, September 26, 2024, 11:30−11:50, Salon 14

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on December 26, 2024

Abstract

Similar to human drivers, it is crucial for Autonomous Driving (AD) systems to comprehend and adhere to traffic rules effectively. This adherence ensures safety and societal acceptance of Autonomous Vehicles (AVs). Integrating traffic rules into the learning process and real-time monitoring of rule adherence are therefore essential for the successful implementation and widespread adoption of AD technologies. To address the need for traffic rule conformance analysis, we introduce a framework based on BARK, our in-house developed scenario simulator. Within this framework, we formalize rules using Linear Temporal Logic (LTL) formulas and employ the quantitative semantics of Signal Temporal Logic (STL) to assess rule compliance with finer granularity. These rules are then integrated into the learning process of Deep Reinforcement Learning (DRL) agents. By conducting experiments across various simulated scenarios within the framework, as well as on a public dataset, we underscore the importance of integrating both traffic rules and quantitative semantics of STL into the development of behavior planner agents. This integration enhances their efficacy and safety in real-world applications.