Key points are not available for this paper at this time.
Aiming at the problem of autonomous path planning for police drones in complex and dynamic environments, this research innovatively proposes an intelligent decision-making method for police drones based on an improved deep Q-network algorithm. This method first adopts the gravity-sensing strategy and the safety counting mechanism to optimize the deep Q-network algorithm. By simulating the gravitational and repulsive relationship between obstacles and target points, it effectively solves the problem of blind exploration in the early training stage of traditional deep Q-network algorithms. Secondly, to address the local circular obstacle problem in partially observable Markov decision, the deep Q-network is further combined, and the long short-term memory network structure is utilized to enhance the model's ability to extract historical state information. The experimental results show that in a static environment, the average number of path steps of the research method is 51.7 steps, and the average success rate of the task is 95.6%. In the dynamic environment, its obstacle avoidance success rate reaches 92.8%, and the path planning time is controlled at 53.4 ms. From this, the proposed method significantly improves the recognition and response capabilities of drones to dynamic targets and obstacles in a local observation environment, thereby reducing path repetition and failure. The research results provide a new idea for constructing an intelligent decision-making system for police drones with high robustness and real-time response capabilities.
Xing et al. (Mon,) studied this question.