SARAD: LLM-Based Safety-Aware Hybrid Reinforcement Learning with Collision Prediction for Autonomous Driving
Kangyu Wu, Peng Cui, Guoxi Chen, Ya Zhang
Why It Matters
What makes this one worth your time
This research addresses critical safety challenges in autonomous driving, potentially leading to more reliable and efficient decision-making systems in real-world applications.
SARAD innovatively merges LLMs with DRL to improve safety in autonomous driving.
Summary
The paper introduces SARAD, a hybrid framework that combines Large Language Models (LLMs) and Deep Reinforcement Learning (DRL) to enhance safety and efficiency in autonomous driving by replacing random exploration with LLM-guided decisions and incorporating a collision prediction module.
Key contributions
- Development of a safety-aware hybrid framework combining LLMs and DRL.
- Introduction of a collision predictor module fine-tuned with historical data.
- Implementation of an attention discriminator to enhance DRL policy optimization.
Notable insights
- The use of Retrieval-Augmented Generation (RAG) to enhance decision-making in DRL is a novel approach that leverages expert knowledge effectively.
- Integrating an attention discriminator for policy optimization could lead to more informed and safer driving strategies.
Possible limitations
- Not stated in the abstract.
Abstract
arXiv:2605.28583v1 Announce Type: cross Abstract: Ensuring both safety and efficiency in decision-making for autonomous driving systems remains a fundamental challenge. Traditional Deep Reinforcement Learning (DRL) suffers from unsafe random exploration and slow convergence, while Large Language Models (LLMs) demonstrate inherent latency in real-time inference operations. To address these limitations, this paper proposes SARAD, a novel safety-aware hybrid framework that synergizes LLMs and DRL for autonomous driving. SARAD substitutes the random exploration of DRL with Retrieval-Augmented Generation (RAG)-enhanced, LLM-guided decisions sourced from a dynamic expert knowledge repository. An attention discriminator is proposed to integrate the prior knowledge of LLMs into DRL policy optimization. A collision predictor module, fine-tuned with historical collision data, is further designed to improve vehicle safety. Extensive experiments show that SARAD achieves significant performance improvements in the Highway-Env simulator, validating the effectiveness of the proposed model in autonomous driving.