Enhancing the Robustness of UAV Search Path Planning Based on Deep Reinforcement Learning for Complex Disaster Scenarios

Le résumé fourni par la source

In disaster scenarios, unmanned aerial vehicle (UAV) rescue path search missions have become an increasingly critical issue. This paper, proposes a Double Deep Q-Network State Splitting Q-Network (DDQN-SSQN) path planning algorithm, which planning rescue route guiding UAV to maximize the detection of ground-trapped devices by analyzing variations of communication signal strengths. However, in the post-disaster environment, communication signal strengths are vulnerable disturbed that severely compromise DDQN-SSQN reliability. To address this problem, in the first stage, we propose a novel State-and-Reward Joint Attack method (Adv-SR) that simulates electromagnetic perturbation to communication signal strengths while incorporating an action detection model to identify algorithmic vulnerabilities, realize attacks on both state space and reward space of DDQN-SSQN. In the second stage, to enhance the robustness of DDQN-SSQN against electromagnetic perturbation attacks, we propose a State-and-Reward Joint Defense method (Def-SR), using dual prediction networks to restore disturbed space. Simulation results demonstrate that compared with traditional attack methods, Adv-SR reduces the task success rate to 12.6% while decreasing attack frequency to 16.67%. Moreover, Def-SR significantly improves the robustness of DDQN-SSQN, restoring the task success rate to 98.4% under attack, showing superior performance compared to single-state or single-reward defense methods. Moreover, Def-SR significantly improves the robustness of DDQN-SSQN, restoring task success rates to 98.4%, showing superior performance compared to single-state or reward defense methods.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.