Abstract

This paper focuses on the problem of emergency restricted zone avoidance for high-speed drones. Existing path planning methods struggle with inefffciency due to large search spaces, slow convergence rates, and difffculties in path planning for high-speed drones. To overcome these challenges, we propose a novel approach that integrates rules with deep learning. This hybrid approach simpliffes decision-making by converting temporal decisions into a limited set of middle way-points, signiffcantly reducing the complexity of both the state and solution search space. Additionally, rules are employed to prevent aimless exploration within the solution space. To enhance the algorithm performance, we introduce a situation prediction model, which is trained to capture the relationship between way-points and ffight outcomes, such as restricted zone encounters and energy consumption. Experimental results demonstrate notable improvements over purely rule-based methods, with high success rates in avoiding restricted zones and maintaining sufffcient kinetic energy to reach the goal. This approach effectively addresses the challenges posed by high-speed drones operating under complex physical models and dynamic emergency scenarios.

Statistics