Integrated V2X communications and reinforcement learning for bus operational control strategy and passenger experience

Abstract Bus bunching remains a critical challenge in urban transit systems, leading to service unreliability, increased passenger wait times, and operational inefficiencies. While existing solutions often address components of this problem in isolation—such as headway-based holding or transit signal priority—they lack a holistic approach that integrates real-time data with adaptive, passenger-centric optimization. This paper introduces the bus operational control strategy (BOCS), a novel integrated framework that synthesizes comprehensive vehicle-to-everything (V2X) communications (V2V, V2I, V2P) with advanced reinforcement learning (RL) to mitigate bunching through real-time adaptive multi-objective optimization. BOCS distinguishes itself from existing frameworks through three unique contributions: (1) a comprehensive V2X data layer that provides unprecedented situational awareness beyond conventional V2I-only systems, (2) a dynamically weighted multi-objective optimization function with adaptive normalization that balances headway adherence with passenger experience metrics in real time, and (3) a hybrid RL architecture using Soft Actor-Critic algorithm with continuous action spaces and physics-informed state representations. The framework was validated through high-fidelity microsimulation with fifty independent replications across two real-world bus routes (Routes 35 and 37) from Gainesville Regional Transit System and two additional simulated routes representing diverse operational contexts. Statistical analysis using paired t-tests and Cohen’s d effect size calculations confirmed significant improvements: BOCS achieved 65–71 per cent reduction in headway deviation (P < 0.001, d = 2.15), 29 per cent reduction in average waiting time (P < 0.001, d = 1.87), and 51 per cent reduction in overcrowding (P < 0.001, d = 1.92) compared to robust headway-based controllers, Q-learning approaches, and multi-agent deep deterministic policy gradient methods. Comprehensive sensitivity analysis demonstrated robustness to V2X communication degradation (maintaining >80 per cent performance at 75 per cent V2X reliability) and generalizability across diverse demand patterns. Ablation studies quantified the contribution of each V2X component, revealing that V2I provides the foundation (72 per cent performance), while V2V and V2P integration enhances performance to 92 per cent. Computational analysis confirmed real-time feasibility with average decision latency of 128 ms per control cycle, well within operational requirements. These results confirm that the integration of comprehensive connectivity, adaptive artificial intelligence, and human-centric design is fundamental to advancing resilient and efficient public transit systems.