基于强化学习的四旋翼高度控制器设计  

Design of Quadrotor Flight Controller Based on Reinforcement Learning

作　　者：苏立军[1] 朱红娟[2] 齐晓慧[1] 董海瑞[1] 

机构地区：[1]军械工程学院无人机工程系,河北石家庄050003 [2]河北政法职业学院计算机系,河北石家庄050061

出　　处：《测控技术》2016年第10期51-53,58,共4页Measurement & Control Technology

摘　　要：针对四旋翼飞行器高度控制通道易受干扰的问题,在先验飞行数据基础上,利用局部加权线性回归算法建立了四旋翼飞行器的非参数模型,该模型考虑了实际飞行中干扰因素对高度控制的影响。以此非参数模型为基础,运用强化学习算法构建了四旋翼飞行器高度控制算法,并以离线方式优化了控制算法的结构参数。最后通过实际的飞行实验验证了控制算法的合理性。To solve the problem that the quadrotor aircraft height control channel is susceptible to be interfered, on the basis of a priori flight data, a non-parametric model of quadrotor aircraft is established by using the locally weighted linear regression algorithm, which considers the influence of interference factors on the height control in actual flight. Based on this non-parametric model, a quadrotor aircraft height control algorithm is constructed by using reinforcement learning algorithm and the structural parameters of the control algorithm is optimized in offline mode. Finally, the actual flight experiment validated the control algorithm.

关 键 词：四旋翼飞行器 局部加权线性回归 非参数模型 强化学习 

分 类 号：TP15[自动化与计算机技术—控制理论与控制工程]