结合跳点引导的无人机随机搜索避撞决策方法

针对无人机在城市空域环境和密集交通流下的避撞决策问题，提出马尔科夫决策过程（MDP）和蒙特卡洛树搜索（MCTS）算法对该问题进行建模求解。蒙特卡洛树搜索算法在求解过程中为保证实时性而使其搜索深度受限，容易陷入局部最优，导致在含有静态障碍的场景中无法实现避撞的同时保证全局航迹最优。因此结合跳点搜索算法在全局规划上的优势，建立离散路径点引导无人机并改进奖励函数来权衡飞行路线，在进行动态避撞的同时实现对静态障碍的全局避撞。经过多个实验场景仿真，其结果表明改进后的算法均能在不同场景中获得更好的性能表现。特别是在凹形限飞区空域仿真模型中，改进后的算法相对于原始的蒙特卡洛树搜索算法，其冲突概率降低了36%并且飞行时间缩短47.8%。

Collision avoidance decision method for UAVs in random search combined with jump point guidance

Aiming at the collision avoidance decision of UAVs in urban airspace environment and dense air traffic flow, Markov Decision Process (MDP) and Monte Carlo Tree Search (MCTS) algorithms are proposed to model and solve this problem. To ensure real-time performance, the MCTS algorithm is limited in search depth, thus easy to fall into the local optimal solution, resulting in the inability to achieve collision avoidance and track optimization in the scene with static obstacles. Therefore, a discrete path point combining the advantages of Jump Point Search algorithm in global planning is established to guide UAVs, and the reward function is improved to balance the flight path, avoiding local dynamic collisions and static obstacles simultaneously. The simulation results show that the improved algorithm can achieve better performance in different experimental scenarios, particularly in the simulation model of concave restricted flight area, where, compared with the original MCTS algorithm, the collision probability of the improved algorithm and the flight time are reduced by 36% and 47.8%, respectively.