Research on Task Offloading and Delay Optimization of Autonomous Aerial Vehicle-Assisted Medical Edge Computing Based on Deep Reinforcement Learning

Integrating autonomous aerial vehicle (AAV) edge computing with medical systems enhances real-time data processing, but any significant delays could hinder timely medical responses. Additionally, there are technical challenges in optimizing task offloading within high-dimensional continuous action spaces. To address this issue, we propose the Dual Delayed Deep Deterministic Policy Gradient with Behavior Cloning and Softmax (TD3-BC-S) optimization algorithm based on Deep Reinforcement Learning (DRL), which designs optimal computation task offloading strategies within high-dimensional states and continuous action spaces. This algorithm utilizes a dual Q-network to minimize overestimation and combines behavior cloning for stable policy learning. Through extensive simulation results, we demonstrate the superior performance gain of the proposed TD3-BC-S algorithm compared with traditional algorithms in terms of significantly reducing offloading latency and enhancing computational efficiency, thereby improving the overall effectiveness and safety of medical services, particularly during public health emergencies.