Research Of Mechanical Arm Action Learning From Demonstration Strategy Based On Neural Network

In recent years,robots have been widely us ed in all aspects of our so cial production and life,especially service robots.Compared with the industrial robots that perform fixed actions,service robots need to share working space with people and the operation of the mechanical arm is complex and changeable.Therefore,in order to promote robots in a wider range in daily life,robots need to have the ability of self-learning.Learning from demonstration is to apply the imitation principle to the robot's action learning,which can im prove the efficiency and intelligence of robot learning.Therefore,this paper researches the problem of robot action teaching and learning,focusing on the combination of neural network technology and learning from demonstration mechanism to solve the problem of robot action learning.The main work of this pape r is as follows:Firstly,a manipulator action learning system based on learning from demonstration mechanism is constructed.According to the process of learning from demonstration of the manipulator action,the architecture of the manipulator action learning system based on the learning from demonstration mechanism is established,and the model of the manipulator arm system was constructed;Based on the established system model,the relevant state collection of the teaching information of the manipulator are implemented.Secondly,the improved BP neural network is constructed based on the teaching information for learning from action demonstration.To describe the learning from demonstration strategy of the mechanical arm,it is the rotation angle of each joint angle of the mechanical arm;BP neural network is constructed based on the mapping relationship between mechanical arm teaching information to learn the strategy,in order to improve the generalization performan ce of the network,the BP neural network is optimized by Bayesian regularization algorithm;Finally,the effectiveness of the network is verified by experiments.Thirdly,a learning from demonstration model based on SDA-RBF neural network is proposed.This paper introduces the existing RBF neur al network model and auto-encoder model,improves the RBF neural network with sparse denoising autoencoder,proposes SDA-RBF neural network;Finally,the validity of SDA-RBF neural network model is tested and compared.Finally,the simulation system of learning from demonstration strategy of the mechanical arm action is established.A mechanical arm simulation model was built in ADAMS simulation software,and the experimental research on the learning from demonstration model of mechanical arm based on neural network is carried out.