SVR Controller for a Biped Robot with a Human-like Gait Subjected to External Sagittal Forces

Abstract

This paper describes the control of a biped robot that uses an SVR (Support Vector Regression) for its balance. The control system was tested subjected to external sagittal pulling and pushing forces. Biped robots have leg link structures similar to the human’s anatomy. To be able to maintain its stability under dynamic situations such robotic systems require good mechanical designs and force sensors to acquire the zero moment point (ZMP). Research in biped robotics has recently had a great surge due to the challenges of the subject and the media impact of famous biped robots like Honda’s. (Vukobratovic, 1990) developed a mathematical model of a biped robot and its method of control. Some research works (Zarrugh & Radcliffe, 1979), (Nakamura et al., 2004), (Jang et al., 2002) have reported the gait of biped robots based on human kinematics data, and a very good study of human body kinematics was done by Winter (Winter, 1990). Because a biped robot is easily knocked down, its stability must be taken into account in its gait design. Zheng (Zheng & Shen, 1990) proposed a method of gait synthesis taking into account the static stability. Chevallereau (Chevallereau et al., 1998) discussed dynamic stability through the analysis of the reaction force between the base of the foot and the ground. Unfortunately the defined trajectory does not assure the satisfaction of the stability restriction. To assure the dynamic stability of a biped robot, Shin (Shin et al., 1990) and Hirai (Hirai et al., 1998) proposed standard methods for gait synthesis based on the zero moment point (ZMP). Basically this method consists of designing a desired ZMP trajectory, duly correcting the movement of the torso to maintain the ZMP trajectory as designed. However, because the change of the ZMP to accommodate the movement of the torso is limited, not all desired ZMP trajectories are possible (Park & Kim, 1998). The ZMP position can be obtained computationally using a model of the robot. However there might be a significant difference between the real and the calculated ZMP due to the difference between the real robot’s physical parameters and its approximated mathematical model. To avoid this error, four force sensors are usually used on each foot to obtain an estimate for the real ZMP.