ARTIFICIAL EVOLUTION OF CONTINUOUS-TIME RECURRENT NEURAL NETWORKS FOR THE CONTROL OF AUTONOMOUS MOBILE ROBOTS

摘要

In the search for the automatons and intelligent systems of the future, there are significant difficulties in capturing and expressing the complex dynamic coupling between system and environment. As systems have become more complex, the manual effort required to control them becomes drastically increased or indeed impossible as decades of research have yet to yield technologies that allow robots to operate successfully in the majority of human environments. Artificial evolution of control networks has the potential to develop sophisticate control systems, bypassing the limitations of human design. This paper describes a Graphics Processing Unit (GPU) accelerated highly parallel approach to the evolutionary process, and presents preliminary results for a simplified case. Avenues of further work are presentation and links are drawn between the many similarities of autonomous mobile robotics and complex manufacturing control systems.