The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #93137

Title:Learning to Coordinate Behaviours on a Four-Legged Robot
Date: 1993
Abstract:An algorithm has been developed elsewhere which enables a six-legged robot to learn how to coordinate its leg movements and walk with a statically stable gait. In the current project, the applicability of this algorithm to a four-legged robot was investigated. A physical machine was constructed, along with a computer simulation. There was insufficient time between the completion of the robot and producing this report to test the algorithm on the physical machine, so the results described apply to the simulation only. The basic finding was that the robot could not learn how to coordinate its legs so that it never fell over. Several variations and extensions of the basic algorithm were also investigated, some of which resulted in a marked improvement in performance so that, in some cases, the robot would only fall occasionally. To model the real robot more closely, some trials on the simulation included a degree of noise in the leg movements. This was found to have a (sometimes drastically) detrimental effect on the level of performance achieved. Some reasons for the failure of the algorithm to satisfactorily translate from a six-legged robot to a four-legged robot are discussed, and possible extensions, which may lead to improved performance, are suggested.

[Search These Pages] [DAI Home Page] [Comment]