基于内在动机的强化学习算法在两轮机器人中的研究  

Reseach on Reinforcement Learning Algorithm Based on Intrinsic Motivation for Two-wheeled Robot

作　　者：任红格[1] 向迎帆 李福进[1] 刘伟民[1] 

机构地区：[1]河北联合大学电气工程学院,河北唐山063009

出　　处：《计算机测量与控制》2015年第9期3185-3187,3191,共4页Computer Measurement &Control

基　　金：国家自然科学基金(61203343);河北省自然基金(E2014209106)

摘　　要：针对两轮自平衡机器人在学习过程中遇到的主动性差和以往强化学习对单步学习效率低的问题,受心理学中内在动机理论的启发,提出一种基于内在动机的强化学习算法;该算法利用内在动机信号作为内部奖励,模拟人类心理认知机理并与外部信号一起作用于整个学习过程,提高了智能体的自学习能力,同时采用自组织神经网络进行训练,保证了算法的快速性;通过无扰动和有扰动两种仿真实验的对比,验证了基于内在动机的强化学习算法能够使两轮机器人在未知环境下通过自主学习最终达到平衡,且体现了该算法的鲁棒性和可行性。Aiming at the two-wheeled self-balancing robot in the learning process encountered less-initiative and reinforcement learning to step low learning efficiency in the past,inspired by the intrinsic motivation theory from the psychology,this paper proposes a reinforcement learning algorithm based on intrinsic motivation.This algorithm uses the intrinsic motivation signal as the internal reward,then simulats human psychological nechanism,and applies to the whole learning process with the external signal.That can improve the learning ability.At the same time,by using self-organizing neural network for training,which ensures the rapidity of the system.The undisturbed and disturbed simulation experiment results prove that the reinforcement learning algorithm based on intrinsic motivation can solve the problem of autonomous learning of two-wheeled robot balance control in an unknown environment,and reflects the effectiveness and robustness of the system.

关 键 词：内在动机 强化学习 平衡控制 鲁棒性 两轮机器人 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]