ROAD: Responsibility-Oriented Reward Design for Reinforcement Learning in Autonomous Driving
Authors:
- Yongming ChenYCYongming Chen
- Liewen LiaoLLLiewen Liao
- Hengrui ZhangHZHengrui Zhang
- Yuchen XiYXYuchen Xi
- Songan ZhangSZSongan Zhang
Preprint
Abstract
Reinforcement learning (RL) in autonomous driving employs a trial-and-error mechanism, enhancing robustness in unpredictable environments. However, crafting effective reward functions remains challenging, as conventional approaches rely heavily on manual design and demonstrate limited efficacy in complex scenarios. To address this issue, this study introduces a responsibility-oriented reward function that explicitly incorporates traffic regulations into the RL framework. Specifically, we introduced a Traffic Regulation Knowledge Graph and leveraged Vision-Language Models alongside Retrieval-Augmented Generation techniques to automate reward assignment. This integration guides agents to adhere strictly to traffic laws, thus minimizing rule violations and optimizing decision-making performance in diverse driving conditions. Experimental validations demonstrate that the proposed methodology significantly improves the accuracy of assigning accident responsibilities and effectively reduces the agent's liability in traffic incidents.Keywords
All Related Versions
This publication has 15 references indexed in Scilit:
- VLM-RL: A unified vision language models and reinforcement learning framework for safe autonomous drivingTransportation Research Part C: Emerging Technologies, 2025
- IBD: Alleviating Hallucinations in Large Vision-Language Models via Image-Biased Decoding2025
- LORD: Large Models Based Opposite Reward Design for Autonomous Driving2025
- Evolutionary End-to-End Autonomous Driving Model With Continuous-Time Neural NetworksIEEE/ASME Transactions on Mechatronics, 2024
- Imitation Is Not Enough: Robustifying Imitation with Reinforcement Learning for Challenging Driving Scenarios2023
- Reward (Mis)design for autonomous drivingArtificial Intelligence, 2022
- MetaDrive: Composing Diverse Driving Scenarios for Generalizable Reinforcement LearningIEEE Transactions on Pattern Analysis and Machine Intelligence, 2022
- Ethical dilemmas are really important to potential adopters of autonomous vehiclesEthics and Information Technology, 2021
- Ethical issues in focus by the autonomous vehicles industryTransport Reviews, 2021
- A Reinforcement Learning-Based Adaptive Path Tracking Approach for Autonomous DrivingIEEE Transactions on Vehicular Technology, 2020