ROSA-RL: Uncertainty-Aware Roundabout Optimized Speed Advisory with Reinforcement Learning

ROSA-RL is a newly introduced framework that employs reinforcement learning to provide uncertainty-aware speed advisories for roundabout entry, addressing the challenges posed by mixed traffic scenarios. Utilizing a Transformer-based model, it predicts conflict zone occupancy over a five-second horizon while accounting for multi-agent interactions and uncertainty in future motion and intent. In simulations based on real-world data, ROSA-RL demonstrated superior performance compared to a model-based baseline, enhancing both traffic efficiency and safety for automated and human-driven vehicles.