Optimal Tracking Control Based on Integral Reinforcement Learning for An Underactuated Drone

作     者：Shaobao Li Petar Durdevic Zhenyu Yang 

作者机构：Department of Energy Technology Aalborg University Esbjerg Campus Niels Bohus Vej 8 Esbjerg 6700 Denmark 

出 版 物：《IFAC-PapersOnLine》 

年 卷 期：2019年第52卷第8期

页      面：55-60页

主　　题：Reinforcement learning optimal control neural network inner-outer loop control 

摘      要：A drone is desirable to perform various flying missions with different loads while always guaranteeing optimal flying performance. In this paper, an integral reinforcement learning algorithm is developed for a drone such that it can learn optimal control policy online. The drone is described by an underactuated nonlinear model and the inner-outer loop control strategy is applied for the navigation control. In the outer loop an optimal controller is designed to minimize a cost function with input saturation, and a policy iteration based integral reinforcement learning (IRL) algorithm is proposed. Critic-actor neural networks (NNs) are further applied for online implementation of the IRL algorithm. In the inner loop a quaternion based feedback attitude controller is designed to guarantee system stability. A simulation study is finally provided to demonstrate the effectiveness of the proposed IRL algorithm.