【作者】 王晓芳； 尹依伊；

【Author】 Xiaofang Wang;Yiyi Yin;School of Aerospace Engineering, Beijing Institute of Technology;Beijing Institute of Electronic System Engineering;

【机构】 北京理工大学宇航学院； 北京电子工程总体研究所；

【摘要】 针对多飞行器编队控制问题,在考虑环境中不确定因素的情况下,基于深度强化学习理论设计了一种能够实现飞行器间自主避碰的编队控制方法。基于飞行器定高飞行假设建立了二维编队相对运动模型,将队形变换过程划分为三个阶段,并建立各阶段的马尔科夫决策模型和深度强化学习编队控制网络;设计从弹坐标变换方法,使得基于某一确定领从初始相对位置训练得到的编队控制器能够适用于具有不同初始相对位置的领从飞行器;针对队形变化过程中的碰撞避免问题,设计了飞行器自主避碰策略。最后,对不同初始状态的飞行器及存在碰撞可能性的飞行过程开展仿真计算,验证了本文控制方法的有效性。更多还原

【Abstract】 To solve the problem of multi-flights formation control with the uncertain factors in the environment, a formation control method based on the theory of deep reinforcement learning is proposed, which can also guarantee the collision avoidance between flights. Based on the hypothesis of the fix altitude of the flights, a two-dimensional formation relative motion model is established. The formation changing process is divided into three separated parts, and the Markov model as well as the deep reinforcement learning formation control network of each part is established; The coordinate transformation method is designed so that the formation controller trained based on a certain initial relative position can be applied to flights with different initial relative positions; An autonomous collision avoidance strategy is designed for the collision avoidance during the formation conversion process as well. Finally, simulations are carried out for the aircraft in different initial states with the possibility of collision, which verifies the effectiveness of the control method..更多还原