Research on Intelligent Evasion Methods for UAV Based on Deep Reinforcement Learning

To address the issue of unmanned aerial vehicle (UAV) autonomously evading aerial incoming target, this paper proposes an intelligent evasion method for the UA V based on Soft Actor-Critic (SAC) algorithm. Given the state information of the UA V and the aerial incoming target as input, the proposed method can generate control commands for the UA V as output, achieving end-to-end autonomous evasion decision-making. Based on the evasion model proposed in this paper, we built the air combat environment. This paper introduces a novel reward function used for generating autonomous evasion strategies for UAV, taking into account the situational information of both the UA V and the aerial incoming target. Finally, by comparing the training and simulation results with the Deep Deterministic Policy Gradient (DDPG) algorithm, the paper validates that the intelligent evasion method based on SAC algorithm converges faster, exhibits superior performance, and learns more flexible and intelligent strategy.