Joint Optimization of Packet Survivability and Aerodynamic Energy for Dynamic UAV Activation in VANETs via Deep...

UAV-assisted VANETs are a key component of the 6G vision, yet their practical deployment is hindered by the fundamental conflict between network Quality of Service (QoS) and the high aerodynamic power required for rotary-wing flight. This paper proposes SAVIOR (Survivable Aerial-Vehicular Intelligent Optimization and Routing), a Deep Reinforcement Learning (DRL) framework that jointly optimizes multi-UAV activation and packet routing. Unlike existing approaches that rely on oversimplified linear energy models, SAVIOR integrates a rigorous three-component aerodynamic power model and introduces an M/M/1 queuing-based Survivability Score (S-score) to explicitly quantify packet delivery before Time-to-Live (TTL) expiration. Through a high-fidelity co-simulation using SUMO and Python-TraCI, the SAVIOR agent is able to handle stress-test situations where network demand is higher than capacity (ρ>1.0). A comparative analysis shows that SAVIOR is Pareto-optimal, with a total reward that is 65% higher than that of a static energy-saving policy and a survivability that is 24% higher. Crucially, compared to a performance-maximizing greedy policy, SAVIOR maintains comparable safety-critical QoS while reducing total energy consumption by 19.8%, thereby preventing premature battery depletion and mitigating co-channel interference.