无人机自主引导跟踪与避障的近端策略优化

针对无人机地面动态目标跟踪问题，建立了远距离自主引导与近距离伴飞避障2个阶段的马尔可夫决策过程模型。在此基础上，提出了一种改进的近端策略优化（PPO）算法。考虑到无人机接收到的数据具有时序性且环境状态存在上下文关联，所提算法采用长短期记忆（LSTM）网络，通过无人机与目标的实时位置关系等状态信息来计算奖励值，更新网络参数，并进行自适应优化迭代。通过基于ROS系统的仿真测试平台进行试验，结果表明：所提算法安全有效地实现了侦察任务全过程的自主机动，与传统的PPO算法相比，LSTM的引入缩短了模型训练时间，跟踪与避障的效率明显提高，进一步加强了算法的鲁棒性、准确性和实时性。

Proximal policy optimization for UAV autonomous guidance,tracking and obstacle avoidance

We established a Markov decision process model with two stages of long-distance autonomous guidance and short-distance companion flight avoidance for multi-rotor UAVs to track dynamic ground targets. An improved proximal policy optimization (PPO) algorithm is proposed. Considering that the data received by the UAV are time-sequential and that the environment has contextual relevance, the algorithm uses long short-term memory (LSTM) network to calculate reward values, update network parameters, and perform adaptive optimization iterations through status information such as the real-time position relationship between the UAV and the target. Experiments were conducted with a simulation testing platform based on ROS. Results show that the method proposed safely and effectively realizes autonomous maneuvering during the whole process of the reconnaissance mission. Compared with the traditional PPO algorithm, the algorithm proposed shortens the model training time due to the introduction of LSTM neural network, thus significantly improving the efficiency of obstacle tracking and avoidance. This result further strengthens the robustness, accuracy, and real-time performance of the algorithm.