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

Researchers have introduced a reinforcement-learning system designed to guide vehicles through roundabouts in mixed traffic by forecasting uncertainty in human driving behavior. The system, called ROSA-RL, uses a Transformer-based model to predict whether the conflict zone will be occupied within a five-second window, according to a paper posted to arXiv on June 15, 2026 [1][2]. Roundabouts present a persistent challenge for automated driving systems because human drivers behave in heterogeneous and non-deterministic ways, creating uncertainty about whether the conflict zone will be blocked at the moment a vehicle attempts to enter [2]. The new framework, named Uncertainty-Aware Roundabout Optimized Speed Advisory with Reinforcement Learning, or ROSA-RL, addresses this by coupling probabilistic conflict forecasting with a classical reinforcement-learning controller [2]. A Transformer-based model predicts conflict zone occupancy over a five-second horizon, capturing multi-agent interactions to anticipate upcoming conflicts and available gaps [2]. The prediction outputs encode uncertainty in future motion and intent, and those outputs augment the state of the reinforcement-learning framework, enabling uncertainty-aware speed coordination [2]. In simulations grounded in real-world data, ROSA-RL outperformed a comparable model-based baseline and narrowed the performance gap relative to an ideal setting that assumes fully known occupancy, while improving both traffic efficiency and safety [2]. The source code has been made publicly available on GitHub under the urbanAIthi organization [2]. The paper appeared on arXiv, the open-access e-print repository that hosts preprints across physics, computer science, mathematics, and related fields [6]. arXiv, which began in 1991, passed the two-million-article milestone by the end of 2021 and currently receives roughly 24,000 submissions per month [6]. The repository is not peer-reviewed; papers are approved for posting after moderation [6]. The ROSA-RL preprint was submitted under the Computer Science category with a primary classification of Artificial Intelligence [1]. arXiv also supports a community-collaboration framework called arXivLabs, which allows third-party developers to build experimental tools that appear on article record pages [4]. The framework was formalized in 2020 to enable innovative collaborations while requiring partners to adhere to arXiv’s values of openness, community, excellence, and user-data privacy [4]. Current arXivLabs integrations include the Bibliographic Explorer, which displays citation-tree information, and the CORE Recommender, which surfaces relevant open-access papers from a global network of repositories [5]. The ROSA-RL abstract page includes links to several of these tools, including Connected Papers and Litmaps, which help researchers explore related literature through interactive citation maps [1][5].