Joint Angle Torque generation based on Machine learning Approaches For Humanoid Locomotion

Abstract

How to generate the suitable torques for stable trajectories for Humanoid locomotion is the critical research problem. To analyze this problem first solution was to adopt the conventional method of inverse dynamics, but this method does not provide the unique solution and it is also computationally complex. So, in this research, a novel approach has been proposed with the help of forward dynamics and K-nearest neighbour (KNN) algorithm to find the suitable joint torques. This method also suitable for finding the joint angle torques of the sub-phases of the gait cycle. Hence,The database has been created by dividing the gait cycle in to sub phases and finding the position (θ), velocity(θ˙)  and acceleration (θ¨)  by given the joint angle torques of stable walking. This database has been used in the KNN algorithm for finding the suitable joint angle torques for the given position, velocity and acceleration. In this way, we propose an alternative way to calculate joint torques in a computational efficient manner. In this research, we also proposed a novel method to extract the Homogeneous kinematics transformation matrix for every joint with help of energy silhouettes of 3D subjects. The hierarchical kinematics synthesis is used to synthesis the approach for articulated skeleton which generates from 3D model. In this context, a full body joint angle trajectories movement articulated in the skeleton form, and key poses are extracted. These key poses are synthesized by the hierarchical kinematics synthesis which includes the revolute and spherical joints corresponding five serial chain of human body. The result shows the process effectively recover the joint angle trajectories of the full body articulated skeleton.