Robot dynamics modeling method based on deep learning

Abstract

The invention discloses a robot dynamics modeling method based on deep learning and belongs to the field of intelligent robots. Data are acquired and divided into a training set and a data set, and adynamics model is a built, and a recurrent neural network (RNN) is constructed; the training set is divided according to the time step and is input into an input hidden layer, and is converted into three-dimensional data to reach a GRU cell layer, currently input information is combined with previous information, and the proportion of state information, participated into a newly generated state, at the previous moment is calculated; and then a current candidate state obtained due to calculation and information of the previous time step moment are selected through an updating gate, a hidden layer state at the current moment is obtained, transmitted to a next time step, and output to an output hidden layer, and an acquired real result with a predicted value smaller than or equal to an errorthreshold is obtained and is an optimal value. Finally, the data set is utilized for detecting a gated recurrent unit (GRU) network. According to the method, the torque detecting precision is improved, the training time of an input signal is greatly shortened, and the gradient error of traditional counterpropagation is reduced.

Description

technical field [0001] The invention belongs to the field of intelligent robots, in particular to a robot dynamics modeling method based on deep learning. Background technique [0002] An important application of the robot dynamic model is to control the robot. The torque required for the robot movement can be accurately calculated through the dynamic equation; however, due to the influence of factors such as disturbance, elasticity, nonlinear friction, and load changes, the parameters of many dynamic models It is difficult to be determined. Traditional dynamics methods (such as Lagrangian, Newton Euler, and Kane) are difficult to accurately model the dynamics of robots, which cannot meet accurate practical applications. [0003] In recent years, the development of deep learning has ushered in a turning point in the solution of this problem; the neural network has a strong nonlinear mapping ability, and by training a certain amount of data, it is not necessary to artificiall...

AI Technical Summary

Problems solved by technology

However, the ESN machine learning algorithm does not fully consider the characteristics of the input signal, usually ignoring the output feedback connection

Therefore, the echo state characteristics of ESN cannot be fully satisfied within a certain period of time, and cannot provide higher accuracy and faster convergence speed for time series prediction

Images