深度强化学习在室内无人机目标搜索中的应用

针对室内无人机随机目标搜索效率不高、准确率低等问题，提出了一种基于空间位置标注的好奇心驱动的深度强化学习方法。用正六边形对探索空间进行区域划分，并标记无人机在各区域的访问次数，将其作为好奇心，产生内部奖励，以鼓励无人机不断探索新领域，有效避免其陷入到局部区域；训练时采用近端策略优化算法（PPO）优化神经网络参数，该算法能使无人机更快找到最优搜索策略，较好躲避障碍物，有效缩短训练周期，提升搜索效率和准确率。

Application of Deep Reinforcement Learning in Indoor UAV Target Search

Inview of the low efficiency and low accuracy of indoor random target search by UAV, this paper proposes the deep reinforcement learning algorithm by curiosity-driven exploration based on spatial location annotation. Firstly, it divides the exploration space by regular hexagon, and records the number of the UAV’s visiting in each single area. Then, it generates the internal rewards by the visiting records, which can encourage the UAV to explore new areas continuously and effectively avoid LUAV sinking into local areas. When training the neural network, it uses PPO（Proximal Policy Optimization） algorithm to optimize the parameters, which can find the optimal search strategy faster, avoid the obstacles better, shorten the training period, and improve the search efficiency and accuracy.