MoSAIC-PPO: Mobility-aware Service Allocation with Integrated Constraints using Proximal Policy Optimization for Vehicular Edge Computing

Vehicular Edge Computing (VEC) enables latency critical vehicular applications by offloading computation from vehicles to nearby edge service providers (ESPs). However, high vehicle mobility, bursty service demand, limited edge resources, and inter-service dependencies make static or re active orchestration strategies ineffective. This paper presents MoSAIC-PPO, a hybrid orchestration framework that jointly optimizes microservice placement, replica scaling, and migration in dynamic VEC environments. MoSAIC-PPO integrates domain specific heuristics with Proximal Policy Optimization (PPO) to balance short-term responsiveness and long-term performance optimization. The heuristic layer incorporates EWMA-based de mand forecasting, safety-stock replica provisioning, dependency aware service placement, and hysteresis-controlled migration to ensure feasibility and stability under fluctuating workloads. The PPO agent refines orchestration decisions over time by optimizing a multi-objective reward function that captures end to-end service latency, migration overhead, cloud offloading ratio, edge resource utilization, and inter-service dependency delay. The framework is evaluated using realistic SUMO-generated vehicular mobility traces from Luxembourg city. Extensive trace driven experiments demonstrate that MoSAIC-PPO consistently outperforms Random, Greedy, heuristics, and DRL baselines in terms of average and P95 latency, SLA violations, cloud offloading, and edge resource utilization across varying ESP densities and vehicular loads.