Research On Imitation Posture Judgment Strategy Of Humanoid Robot Based On Machine Learning

The imitation behavior of the humanoid robot acts as a professional physician in the adjuvant treatment of autism.Completing the rehabilitation training can liberate the doctor from the high-repetition treatment activities,greatly alleviating the contradiction between supply and demand of medical resources,which is of great significance.However,in the process of imitation,the humanoid robot affects the stable state of the robot due to the movement of the leg.The fall of the robot directly leads to the failure of the imitation task and even the damage of the robot.In this paper,the problem of falling over in the process of robot simulation is proposed.The RBF neural network based on simulated annealing adaptive Particle Swarm Optimization algorithm and the improved Multi-Layer Perceptron network algorithm are used to determine whether the imitate posture of the robot.The specific research contents are as follows:Firstly,the imitation experiment platform is built,and the collected data is optimized in real time by the combined filtering algorithm,and the changes of the robot joints' degrees are stored to realize that the robot can repeat what it has learned.At the same time,the sample data of various postures of the robot is preprocessed,to prepare for the next step which to use the algorithm to determine whether the imitation posture falls.Secondly,an algorithm based on simulated annealing algorithm for PSO-RBF neural network is proposed to determine the robot's imitation posture.In order to solve the problem of poor convergence speed and low accuracy of traditional RBF neural network,the PSO algorithm is used to optimize it.At the same time,In order to solve the problem that the classical PSO algorithm is easy to fall into the local optimal value,the simulated annealing PSO algorithm is proposed.Experiments show that the proposed algorithm has higher convergence speed and accuracy than BP neural network,support vector machine SVM and traditional RBF neural network.Finally,in order to improve the accuracy of the imitative attitude determination,an improved MLP deep learning network model is proposed to determine the robot's imitation posture.In order to solve the problem of poor nonlinear representation ability of traditional MLP network model,a dense connection method of neural network layer is proposed.Two improved methods of Dropout and regularization are used to solve the problem that network training is easy to over-fitting.Experiments show that the proposed algorithm has higher accuracy than the RBF neural network method based on simulated annealing adaptive PSO,and effectively avoiding the fall of the robot during the imitation process,ensuring that the robot can safely complete the imitation task.