可生长仿生学习算法在机器人平衡控制中的研究  

Research of Growth Bionic Learning Algorithm in Balance Control on Robot

作　　者：李福进[1] 王志龙[1] 任红格[1] 高建宇[2] 

机构地区：[1]河北联合大学电气工程学院,河北唐山063009 [2]河北联合大学轻工学院,河北唐山063009

出　　处：《计算机测量与控制》2014年第4期1105-1107,共3页Computer Measurement &Control

基　　金：国家自然科学基金项目(61203343)

摘　　要：针对两轮机器人的运动平衡控制问题,提出一种基于Q学习的生长细胞结构(GCS)网络的仿生学习算法;GCS网络除了具有SOM网络的竞争机制外,它还可以通过新神经元的不断生长,自组织地进行演化,Q学习算法是一种无模型强化学习算法,它可以改善学习能力,但是它只适用于状态离散化的控制系统中;将GCS网络的生长特性应用到Q学习算法中,通过网络输出的获胜神经元的信息来优化Q值,实现了状态连续系统的无模型控制,并且在两轮机器人上做了仿真实验;结果表明,当神经元数为12个时,机器人才开始受控,但是机器人本体的倾角振荡角度过大,位移不受控制;当神经元数增加到25个时,机器人本体的倾角在很小的角度范围内波动(大约0.2°),位移大约在0.05m的位置达到平衡,机器人的运动平衡达到了很好的控制效果。According to the motion balance for a two--wheeled robot control problems, this paper puts forward a growing cell structure （GCS） network based on Q learning bionic learning algorithm. GCS network in addition to the competition mechanism with SOM network, and it can also process self organization evolution through the continuous growth of new neurons. Q learning algorithm is a model free rein- forcement learning algorithm, and it can improve the learning ability of the control system, but it is only suitable for the discretization of state. We made the growth characteristics of GCS network to be applied to the Q learning algorithm, and optimized the Q value through the information of the winning neuron which comes from the network. Ultimately, we achieved the model free control of a continuous state sys- tem, and made simulation experiments on two--wheeled robot. The results showed that when the number of neurons was 12, the robot star- ted under control. But the oscillation angle of the robot body was too large and displacement was uncontrolled; When the number of neurons increased to 25, the inclination of the robot fluctuated in a small angle range （about 0.2 °） and displacement achieved balance at the position about 0. 05m. So, the robot＇s movement balance achieved a good control effect.

关 键 词：GCS网络 Q学习 仿生学习 运动平衡 两轮机器人 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]