Beyond Safety Filtering: Control Barrier Function-Informed Reinforcement Learning for Connected and Automated Vehicles

2026-05-16

AI Summary

This work addresses the challenge of designing safe and effective reward functions in multi-agent reinforcement learning for complex traffic scenarios, where handcrafted rewards are difficult to tune and often fail to guarantee safety. The authors propose a novel approach that directly translates control barrier function (CBF) constraint values into explicit safety-aware reward signals, integrating them into the multi-agent reinforcement learning framework as a principled alternative to heuristic reward shaping. Evaluated on a cooperative driving task at a four-way, multi-lane intersection, the method significantly improves both safety and overall performance. Notably, it demonstrates strong robustness across varying reward hyperparameters, consistently achieving stable and superior results without extensive tuning.

Abstract

Reinforcement Learning (RL) uses rewards to guide learning, yet reward design is typically hand-crafted using heuristics that can be difficult to tune. We propose a Control Barrier Function (CBF)-informed reward design for Multi-Agent RL (MARL) that converts CBF constraint values under joint MARL actions into a reward signal that explicitly guides safe learning. We compare against two heuristic reward baselines in a four-way multi-lane intersection with connected and automated vehicles. Results show that our method achieves the highest task performance and is less sensitive to reward hyperparameters, yielding consistently strong performance across the tested hyperparameter range. Code for reproducing the experimental results and a video demonstration are available at https://github.com/bassamlab/SigmaRL.

Problem

• Reinforcement Learning
• Control Barrier Function
• Connected and Automated Vehicles
• Reward Design
• Multi-Agent RL

Innovation

• Control Barrier Function
• Reinforcement Learning
• Safe Learning
• Multi-Agent RL
• Reward Design