| 基于深度强化学习的无人机姿态控制器设计 |
| |
| 引用本文： | 王伟,吴昊,刘鸿勋,杨溢.基于深度强化学习的无人机姿态控制器设计[J].科学技术与工程,2023,23(34):14888-14895. |
| |
| 作者姓名： | 王伟 吴昊 刘鸿勋 杨溢 |
| |
| 作者单位： | 南京信息工程大学 |
| |
| 基金项目： | 江苏省科学技术厅基础研究计划自然科学基金 |
| |
| 摘 要： | 为了能让四旋翼无人机的姿态控制器具有强大的目标值追踪与抗外部干扰的能力,提出了一种基于参考模型的深度确定性策略梯度的四旋翼无人机姿态控制器设计。该方法通过神经网络,将四旋翼无人机的状态直接映射到输出。本文的强化学习算法是结合深度确定性策略(deep deterministic policy gradient,DDPG)和深度神经网络所设计的。在DDPG算法结构中,进一步加入参考模型,规避控制量太大造成的系统超调,增强了系统的稳定性以及鲁棒性。同时,修改了强化学习中奖励的构成,成功消除了系统的稳态误差。经过研究实验表明,该控制方法可以对目标值进行快速地追踪且有着较强的鲁棒性,可见该控制器相比于传统的控制器,提高了其目标值追踪能力以及抗干扰能力。
|
| 关 键 词： | 深度强化学习 姿态控制 神经网络 参考模型 |
| 收稿时间： | 2022/9/24 0:00:00 |
| 修稿时间： | 2023/8/26 0:00:00 |
|
| |
Wang Wei,Wu Hao,Liu Hongxun,Yang Yi.
|
| |
| Authors: | Wang Wei Wu Hao Liu Hongxun Yang Yi |
| |
| Institution: | Nanjing University of Information Science & Technology |
| |
| Abstract: | In order to make the UAV attitude controller have strong ability of target value tracking and anti-external interference, A design of a quadrotor UAV attitude controller based on the deep deterministic policy gradient of the reference model was used this paper proposes. The network directly maps the state of the quadrotor UAV to the output. The reinforcement learning algorithm used in this paper is designed by combining deep deterministic strategy (DDPG) and deep neural network. In the Actor-Critic structure of the DDPG algorithm, a reference model is further added to avoid system overshoot caused by too much control, and enhance the stability and robustness of the system. At the same time, the composition of the reward in reinforcement learning is modified, and the steady-state error of the system is successfully eliminated. The research and experiments show that the control method has strong robustness to the target value tracking and anti-interference. It is concluded that the controller has improved its target value tracking ability and anti-interference ability compared with the traditional controller. |
| |
| Keywords: | deep reinforcement learning attitude control neural network reference model |
|
| 点击此处可从《科学技术与工程》浏览原始摘要信息 |
| 点击此处可从《科学技术与工程》下载免费的PDF全文 |
|
