Adaptive Dynamic Balance of Two and Four Legged Walking Robots

摘要

Two legged and four legged walking are the most versatile forms of landlocomotion in the sense of maneuverability and the ability to traverse irregular terrain. Unfortunately, the problem of practical bipedal walking with dynamic balance has so far eluded solution using classical and other control techniques. We have been investigating the use of neural networks for on-line gait modulation during walking, using a 10 degree of freedom biped robot and a 20 degree of freedom quadruped robot developed in our laboratory. The research in the current project involves extending our adaptive walking control architecture, developed in prior research, to provide good performance over a wider range of stepping lengths and rates, and to provide good stability during both minor and significant deviations from nominal behavior. This should result in a truly practical control structure for walking with dynamic balance. This is a project progress report for the research carried our prior to June, 1996.