摘要无人机因体积小、成本低、机动性强等特点，目前被广泛应用于各类应急事故响应中。为提升未知环境下无人机避障导航的能力，本研究将事件流和深度数据流引入强化学习网络，提出一种基于事件相机和深度相机的近端策略优化（ED-PPO）算法的无人机避障导航方法。首先，利用UE4和AirSim平台搭建虚拟障碍环境，通过AirSim平台模拟事件相机和深度相机进行动态环境感知；其次，结合感知图像和无人机参数设计状态空间、动作空间和奖励函数；最后，基于卷积神经网络提取状态空间的数据特征，并将其输入到近端策略优化（PPO）算法中进行训练。仿真结果表明，ED-PPO算法能够实现无人机在未知环境下的避障导航和航线规划，采用ED-PPO算法训练的模型具有更优的收敛速度和训练效率。

abstractsUnmanned aerial vehicle（UAV）is characterized by small size，low cost and high mobility，etc. Currently，it is widely used in various accidents’ emergency response. In order to improve the UAV's obstacle-avoiding navigation ability in unknown environment，event stream and deep data stream are introduced into reinforcement learning network，and an event and depth camera-based proximal policy optimization（ED-PPO）algorithm is proposed for UAV obstacle avoidance navigation. Firstly，UE4 and AirSim were used to build a virtual obstacle environment，and AirSim platform was used to simulate event camera and depth camera for dynamic environment perception. Then，the state space，action space and reward function are designed by combining the perception image and UAV parameters. Later，data features of the state space were extracted based on convolutional neural networks and input into the proximal policy optimization（PPO）algorithm for training. The simulation results show that ED-PPO algorithm can realize the obstacle avoidance navigation and route planning of UAV in unknown environment. The model trained by ED-PPO algorithm has better convergence speed and training efficiency.