A Study of the Brachiation Type of Mobile Robot. 7th Report. Behavior Learning for Hierarchical Behavior-based Controller.

摘要

This paper introduces a brachiation type locomotion robot with 13D. O. F. and proposes a learning algorithm for a hierarchical behavior based controller. The brachiation type locomotion robot "Brachiator III" has similar mechanism and dimensions to a long-armed ape, which can perform dexterous motion using a redundant mechanism. A hierarchical behavior-based approach is useful to design a controller for a dynamical motion with multi-degree of freedoms. However some difficulties remain in adaptation phase when the objective behavior or environment conditions are changed. We propose a novel algorithm to improve a global behavior against these changes using Newton Raphson method. This algorithm measures the effects of local behavior controllers to the global behavior as a gradient information, and changes the amplitude of each behavior controller in fewer trials. This learning algorithm is applied to the brachiation robot control.