Research Of Biped Robot Self-Learning Control Algorithms

The biped robotics is a cross discipline which involves the bionics, multi-rigid body complex dynamics, multi-sensor data fusion and control engineering. And it is one of the most important branches. Bipedal walking mode have the high flexibility, and has much better adaptation to environment in human living-space than wheel type, crawl type and pedrail type mobile robots. And has a wider Potential substitute for human work such as healing service and working in hazard condition. Therefore, the research on the biped robot has not only the higher academic value, but also has comparatively practical significance.In this paper, we develop a simple model of biped robot, which is based no the human's walking course and the human's physical structure. In the D-H coordinate, direct kinematic model and inverse kinematic model and built based on the homogeneous coordinate conversion. And this provides the foundation for gait plan. Walking pattern is designed as follow. Stabilization condition of biped robot is studied in the paper with the method of zero moment point (ZMP). Based on the stability criterion, the walking trajectory of biped robot is attained by using cubic curve interpolation. The method proves to be reasonable and effective based on the simulation results under Matlab.Then introduced the knowledge of support vector machine, support vector regression and fuzzy logic.Based on the characteristics of fuzzy information and uncertainty of sample data in the actual system, a three-dimensional fuzzy support vector regression is proposed in this paper.Through add the fuzzy-dimensional into the traditional support vector regression to deal with uncertainty information existed in the sample data.Based on the Energy-Efficiency, a three-dimensional fuzzy support vector regression is proposed in this paper. The on-line controller is established based on the learning of the date. The sample date is come from Energy-Efficiency controller, Through TH-FSVR we ca resolve the learning problem, when the sample data with fuzzy information and uncertainty information in the actual system Simulation results show the accuracy and robustness of the three-dimensional fuzzy support vector regression is are significantly better than the traditional support vector regression.