基于神经网络前馈补偿的欠驱动机器人越障控制  

Obstacle Avoidance Control of Underactuated Robot Based on Neural Network Feedforward Compensation

作　　者：杨志成[1] 冯豫韬[1] 张利霞[1] 齐华山 倪景秀 YANG Zhi-cheng;FENG Yu-tao;ZHANG Li-xia;QI Hua-shan;NI Jing-xiu(College of Biochemical Engineering, Beijing Union University, Beijing 100023, China)

机构地区：[1]北京联合大学生物化学工程学院,北京100023

出　　处：《测控技术》2017年第11期89-92,97,共5页Measurement & Control Technology

基　　金：国家自然科学基金资助项目(51578065);北京市教委资助项目(KM201611417006);北京联合大学新起点计划项目资助(ZK10201504)

摘　　要：针对欠驱动巡检机器人相对于传统机器人控制量少于控制对象的特点,依靠Lagrange定理建立机器人系统动力学模型和电缆弯曲数学模型。根据障碍物外形特点给出障碍物特征量的提取方法;根据障碍物较大,常规的运动控制输出量不宜变化过大的特点,以及机器人越障过程中要保持机体平衡,依据动力学模型提出了基于神经网络补偿的前馈-反馈控制器,以实现对系统姿态控制和越障控制。控制器具有自学习、自动补偿的能力,对非线性对象有较好的控制作用。仿真结果表明,该方法相对于常规PID控制、普通模糊控制具有超前调节、超调量小、抗扰动能力强的特点。通过实验,验证了基于神经网络补偿的前馈-反馈控制器的合理性。According to the Lagrange theorem, the dynamic model of the robot system and the mathematical model of the cable bending are established for the less-control underactuated inspection robot. Based on the characteristics of obstacle appearance, the method of extracting the obstacle features is given. Because of the obstacle is large, the conventional motion control output should not change too much, and the robot should keep the balance in the process of obstruction, a feedforward-feedback controller based on neural network compensa- tion is proposed according to the dynamic model to realize the system attitude control and obstacle control. The controller has the ability of self-learning and automatic compensation, and has better control effect on nonlinear objects. The simulation results show that the proposed method has the advantages of advanced regulation, small overshoot and strong anti-disturbance ability compared with conventional PID control. The rationality of feedfor- ward-feedback controller based on neural network compensation is verified by experiments.

关 键 词：巡检机器人 欠驱动 神经网络 补偿控制 系统辨识 越障控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置] TH89[自动化与计算机技术—控制科学与工程]