Learning Safe Autonomous Driving Policies Using Predictive Safety Representations
This work addresses safety-critical autonomous driving by enhancing the reward-safety tradeoff, though it is incremental as it extends an existing framework to real-world scenarios.
The paper tackled the challenge of balancing performance and safety in safe reinforcement learning for autonomous driving by testing the SRPL framework on real-world datasets, achieving statistically significant improvements in success rates and cost reductions with effect sizes up to 0.86.
Safe reinforcement learning (SafeRL) is a prominent paradigm for autonomous driving, where agents are required to optimize performance under strict safety requirements. This dual objective creates a fundamental tension, as overly conservative policies limit driving efficiency while aggressive exploration risks safety violations. The Safety Representations for Safer Policy Learning (SRPL) framework addresses this challenge by equipping agents with a predictive model of future constraint violations and has shown promise in controlled environments. This paper investigates whether SRPL extends to real-world autonomous driving scenarios. Systematic experiments on the Waymo Open Motion Dataset (WOMD) and NuPlan demonstrate that SRPL can improve the reward-safety tradeoff, achieving statistically significant improvements in success rate (effect sizes r = 0.65-0.86) and cost reduction (effect sizes r = 0.70-0.83), with p < 0.05 for observed improvements. However, its effectiveness depends on the underlying policy optimizer and the dataset distribution. The results further show that predictive safety representations play a critical role in improving robustness to observation noise. Additionally, in zero-shot cross-dataset evaluation, SRPL-augmented agents demonstrate improved generalization compared to non-SRPL methods. These findings collectively demonstrate the potential of predictive safety representations to strengthen SafeRL for autonomous driving.