Reinforcement learning for mobility and resource aware handover in vehicular networks

Vehicular networks need continuous and reliable connectivity to support safety-critical and latency-sensitive applications. However, fast mobility, rapidly fluctuating wireless conditions, and dynamic edge resource availability make Base Station (BS) handover a complex decision problem. Classical threshold-based methods typically rely on radio measurements and are less effective when mobility context and target-BS resource state must also be considered. As a result, they may cause unnecessary handovers, degraded link quality, and increased delay. Existing learning-based and Mobile Edge Computing (MEC)-aware handover studies have considered radio quality, mobility, and resource factors from different perspectives. However, integrated simulator-level evaluations that jointly expose mobility context, wireless-link quality, and candidate-BS resource state during target-BS selection remain limited. This paper presents a Reinforcement Learning (RL)-based handover scheduler implemented in DriveNetSim that jointly uses radio quality, mobility context, and BS resource-state information for target-BS selection. The scheduler constructs a cross-domain state representation from received signal strength indicator (RSSI), distance, CPU utilization, memory utilization, and aggregate load, and optimizes handover decisions through a multi-objective reward formulation. The proposed scheduler is evaluated in DriveNetSim , a vehicular communication simulator developed for this study to emulate mobility, wireless propagation, and dynamic BS load. The algorithm achieves near-oracle handover behavior, improves median RSSI by 2–8 dB, and maintains reliable signal-to-interference-plus-noise ratio (SINR)-based connectivity under mobility and aggregate interference. It reduces handover frequency by up to 65% and selects target BSs with CPU and memory utilization close to the oracle policy. Over 90% of early decision errors disappear as the policy converges, with long-term regret approaching zero and a 17% improvement in end-to-end delay. These results show that jointly considering mobility, radio quality, and candidate-BS resource state can improve handover reliability, stability, and efficiency in dynamic vehicular environments.