Delay-Sensitive Task Offloading for Emergency Ambulance in Edge-MEC-Cloud Continuum using Deep Reinforcement Learning

In pre-hospital emergency medicine, rapid decisions are critical. AI, 5G, and IoMT can enhance automation, but limited ambulance edge resources, network variability, and changing patient needs in real-time demand intelligent AI task offloading. In such scenarios, dynamic task offloading can optimize realtime computation by considering time sensitivity and resource availability across ambulance edge devices, MEC servers, and cloud servers. This paper presents a patient acuity-aware realtime
offloading framework using Proximal Policy Optimization (PPO) across the edge-MEC-cloud continuum. The offloading problem is modeled as a Markov Decision Process considering patient acuity, deadlines, computation needs, battery, network, and MEC service availability. Actions select offloading locations to minimize delays and maximize task success. Validation with synthetic dataset shows the PPO model outperforms traditional algorithms in delay sensitivity and task completion under resource
constraints.