Roundabouts pose significant challenges for automated driving in mixed traffic due to the heterogeneous and non-deterministic behaviors of human drivers, unknown driving intentions, and the complexity of interactions, leading to uncertainty about the availability of conflict zones upon entry. We present ROSA-RL, an uncertainty-aware Roundabout Optimized Speed Advisory utilizing Reinforcement Learning. This approach facilitates safe and efficient entry into roundabouts for both automated and human-driven vehicles through probabilistic conflict forecasting. A Transformer-based model predicts the occupancy of conflict zones over a five-second horizon, capturing multi-agent interactions to anticipate upcoming conflicts and available gaps. The prediction outputs encode uncertainty in future motions and intents, enhancing the state of a classical RL framework to enable uncertainty-aware speed coordination. Evaluated in simulations grounded in real-world data, ROSA-RL effectively manages uncertainty, outperforming a comparable model-based baseline, and narrowing the gap to an ideal scenario that assumes fully known occupancy while improving traffic efficiency and safety. The source code for this work is available at: github.com/urbanAIthi/ROSA-RL.
Blogger's Review: ROSA-RL effectively addresses the complexities and uncertainties of roundabouts by integrating reinforcement learning with a Transformer model, showcasing the potential of future intelligent transportation systems. Its simulation validation based on real-world data provides robust support for the safety and efficiency of automated driving technologies. The open-source nature of this model will also foster further research and development in related fields.