Computer-controlled human body coordination

[摘要] ENGLISH ABSTRACT:A need for intelligent robotic machines is identified. Research and experiments have focussedon stable, or relatively stable, dynamic simulated systems to demonstrate the feasibility ofembedding advanced AI into dynamic physical systems. This thesis presents an attempt toscale the techniques to a dynamically highly unstable system - the coordination of movementsin a humanoid model. Environmental simulation, articulated systems and artificial intelligencemethods are identified as three essential layers for a complete and unified approach to embeddingAI into robotic machinery. The history of the physics subsystem for this project is discussed,leading to the adoption of the Open Dynamics Engine as the physics simulator of choice. Anapproach to articulated systems is presented along with the EBNF of a hierarchical articulatedsystem that was used to describe the model. A revised form of evolution is presented andjustified. An AI model that makes use of this new evolutionary paradigm is introduced. Avariety of AI variants are defined and simulated. The results of these simulations are presentedand analysed. Based on these results recommendations for future work are made.

[发布机构] Stellenbosch University