摘要
磁导航技术已被广泛应用于室内机器人的自主导航控制中,但是随着工业现场导航路径的交叉多变、磁场耦合的日趋复杂,路径识别的错误率也越来越高.为此,本文针对十字交叉、 X型以及环岛等路径元素构成的复杂导航路径,设计了以英飞凌单片机TC264为核心的导航控制系统.该系统利用磁感应传感器采集的导航路径信息,构建了基于多层感知机(MLP)神经网络的路径识别轻量化模型.并在TensorFlow和keras平台下对模型进行了训练.在部署模型时,将磁感应传感器的输出信号定义为模型的输入信号,将单片机控制舵机的PWM值定义为模型的输出信号.实验数据表明,MLP控制系统识别X型路径的准确率达95%,模型运行一轮的时间为505μs.与传统PID控制算法相比,本文磁导航机器人方向控制系统在路径识别能力和实时性方面都有明显的改善.但是在对MLP神经网路进行训练时,需要不断调整网络参数,才能提高系统的综合控制效果.
Magnetic navigation technology has been widely used in indoor robots to achieve autonomous navigation.However,the path identification error rate increases with the changeable crossing of industrial navigation paths and the more complex magnetic field coupling.For this purpose,a navigation control system based on Infineon microcontroller TC264 for the complex navigation path composed of cross,X and circle island path elements was designed.The system combined the magnetic induction sensor to collect the navigation path information,combined a lightweight model of path recognition based on multilayer perceptron(MLP)neural network,and trained the model on TensorFlow and keras platforms.When deploying the model,the output signal of the magnetic induction sensor was defined as the input signal of the model,and the PWM value of the singlechip controlling the steering gear was defined as the output signal of the model.The experimental data show that the accuracy of the MLP control system to identify the X-type path is as high as 95%,and one round of model running time reaches 505μs.Compared with the traditional PID control algorithm,the magnetic navigation robot direction control system has obvious improvement in path recognition ability and real-time performance.However,when the MLP neural network is trained,the network parameters need to be adjusted continuously to improve the comprehensive control effect of the system.
作者
孙鑫
曲立国
陈国豪
陈晴晴
SUN Xin;QU Liguo;CHEN Guohao;CHEN Qingqing(School of Physics and Electronic Information,Anhui Normal University,Wuhu 241000,China)
出处
《中北大学学报（自然科学版）》
CAS
2022年第6期530-535,共6页
Journal of North University of China(Natural Science Edition)
基金
大学生创新创业训练计划项目(S202110370169)。
关键词
磁导航
多层感知机
神经网络
路径识别
单片机
机器人
magnetic navigation
multilayer perceptron
neural network
path recognition
single chip microcomputer
robot
