Research Of Biped Robot Gait Self-Learning Control Algorithms Based On SVM

Robot technology is the combination of machinery, electronics, materiality and other fields of new technologies. Compared with the wheel type robots, biped robots have the higher capability of obstacles avoidance and environmental suitability, which is also a comprehensive subject concerns with bionics, cybernetics, multi-sensor fusion technology, multi-rigidity-body kinetic computer science and the other technologies, so it has the very high academic value and practical significance.The gait motion planning and learning algorithm of the biped robot were discussed in this paper. By using the simulink software, three points were introduced respectively, a kind of typical gait trajectory planning, two kinds of motion planning under the fell situation, a self-learning algorithm of the motion adjustment in uncertain environment.This paper analyzed the process of human walking, proposed a simplified model of the biped robot. In the D-H coordinate, direct kinematic model and inverse kinematic model were built based on the homogeneous coordinate conversion. Common walking motions and fell to stand up planning were realized through some joints motions were designed with the Zero moment theory and polynomial fitting method, and the approach is verified to be reasonable and effective by Matlab.This paper analyzed the problem of sudden motion during the walking, proposed a control method combined with Support Vector Regression (SVR) and Genetic Algorithm (GA). This method used GA to optimize the three key parameters of SVR to research the most accurate regression model based on the limited samples from walking system; the system was completed by a self-learning close-loop feedback control strategy, which was keeping balance constantly by upper-body motion compensation. At last, several control indexes, such as global stability, Root Mean Square (RMS), calculation speed, were analyzed with the simulink results. It showed that the accuracy and robustness of the GA-SVR are significantly better than the traditional Intelligence Algorithm methods.