Agent-Oriented Evolutionary Control Systems: Determining the Optimal Control Codes for 'Ro', the Oriental Oar

摘要

Defining control systems for previously human performed tasks requires experienced operators working together with engineers and/or programmers, who apply their particular expertise to solve the specific control problem. Yet, this procedure generates human-oriented solutions, causing the defined controller to simulate human conducted control, rather than generate optimized, machine-oriented signals specific for the electromechanical system. Furthermore, many control systems may involve a huge number of inter-dependent variables, incompatible with the familiar four-dimension coordinate visualization, rendering the system incomprehensible for human abstraction and therefore unsuitable for conventional treatment. We present here an alternative, machine-oriented and autonomous control system training model, based on natural evolution, Multi-Agent Systems and Distributed Computing, which is able to profit on modern computer architectures and deal with Evolutionary Search issues in a reduced timeframe. Here we describe the method in practice, as we apply it to solve the "Ro" control problem, a simple one-oar robotic rowing system that allow us to compare the autonomous discovered method to the traditional, human conducted rowing method.