Gait transitions of legged robots and neural networks applications

摘要

In this dissertation, the gait transitions of quadruped and hexapod are studied. Three kinds of gait transition models are developed for both quadruped and hexapod: the gait transition models between two general periodic gaits, from a general periodic gait to the follow-the-leader gait, and from the follow-the-leader gait to a general periodic gait. Solution methodologies for these models are presented and simulation results are shown to demonstrate the applications of these gait transition models. The existence and uniqueness properties of the gait transition model solution and their corresponding gait static stability are analyzed in details. As a specific type of gait transitions, the gait transition models between two wave gaits under constant motion speed are investigated. During the transition, the gait preserves the special characteristic of the wave gait that the two adjacent legs on the same side place and lift at the same time. This special characteristic should provide better stability during the transitions. Six theorems are derived for the gait transition for both quadruped and hexapod under the following three conditions: constant stroke, constant duty factor, and constant cycle time. Proofs to these six theorems are given and simulation results are presented. CMACs are applied to learn four kinds of gait transitions: wave gait to wave gait, wave gait to follow-the-leader gait, walk to trot, and trot to gallop. The learning accuracy and generalization capability of the learned CMACs are evaluated and found to be satisfactory.