Action oriented bayesian learning of the operating space for a humanoid robot

This paper describes a new method of self-modeling based on constructing an operating space for a humanoid robot. This approach takes insights from the pain perception, which is regarded as a measure to ensure self-preservation in nature. The anthropomorphic humanoid robot learns the operating space of his joint actuators and the workspace by using its own movements. In addition, we also propose a new method of path planning which utilizes Rapidly-exploring Random Trees (RRTs) and a probability model which is acquired based on the Gaussian Mixture Model (GMM) and Variational Bayesian (VB) learning for the robot. We also demonstrate that the developed algorithm is robust against dynamical changes in the surrounding environment. Path planning is performed in the joint-angle space for an arm having 5 degrees of freedom (DOFs) by utilizing the proposed method. We conducted several experiments in a real environment in order to verify the advantages of the proposed approach.