Application of differential evolution optimization to robotics and mechanism dimensional synthesis.

摘要

In this research, a recently developed optimization technique Differential Evolution (DE) is applied to robot design and dimensional synthesis of mechanisms. Optimum robot design based on task specifications is studied as a minimization problem, where the joint torque is used as an objective function to be minimized subjected to kinematic, dynamics and structural constraints with design variables being the physical characteristics of links. Obtained results are compared with Simple Genetic Algorithms (SGA) and Genetic Algorithms with Elitism (GAE). The comparison shows that DE requires less number of function evaluations and renders smaller values of objective functions than the other two techniques. The application of DE is extended to the synthesis of four-bar “crank-rocker” type of mechanisms to generate coupler curves and recreates a family of coupler curves with prescribed timing. The special characteristic of DE to extend its search beyond the initially specified bounds combined with a newly developed ‘Geometrical Centroid of Precision Positions’ (GCPP) methodology is suggested as an advantageous alternative to those optimization techniques which highly rely on initial guesses and problem specific information such as gradient and higher order derivatives. The GCPP method is combined with DE and applied to the synthesis of mechanisms for a number of cases with different level of complexity. The initially satisfied accuracy at each precision point is improved by using iterative method of successive optimization.