Continuous locomotion of brachiation robot by behavior phase shift

The authors propose an extension of a hierarchical behavior controller which achieve dynamically dexterous behavior by shifting behavior phase. A controller for a dynamically dexterous behavior is hard to be designed by using any unsupervised learning methods, because of enormous searching space. In order to reduce the searching space and its complexity, a hierarchical behavior structure is effective. We previously proposed a hierarchical behavior controller (Y. Hasegawa and T. Fukuda, 1999), which consists of two kinds of modules: behavior coordinator and behavior controller and adaptation algorithm for scaling of behavior outputs. It is applied to the control problem of a seven-link brachiation robot, which moves dynamically from branch to branch like a gibbon swinging its body. The robot however does not locomote from branch to branch stably by adjusting the amplitude of two behavior controllers. A hybrid adjusting algorithm with amplitude scaling and phase shifting of behavior outputs is proposed. Numerical simulations demonstrate that the obtained controller can successfully generate the stably continuous locomotion