Evolutionary algorithms in kinematic design of robotic systems

Abstract

Proposes an adaptative multi-chromosome evolutionary algorithm (AMEA) to perform task based design of modular robotic systems. The kinematic design is optimized by the AMEA which uses both binary and real encoding (for kinematic and configuration parameters). In the problem considered for illustration, the robotic system consists in a mobile base and a manipulator arm which may be built with serially assembled link and joint modules. The manipulator has to reach a predefined set of goal frames in a 3D cluttered environment. Its design is evaluated with geometric and kinematic performance measures. Optimization results for a 3D task are given and compared with a simple genetic algorithm. They clearly show the superiority of the multi-chromosome representation and adaptive operators in term of computing time and criteria optimization performance.