Genetic Approach for Coupled Dynamics Optimization in a Multiple Degree-of-Freedom System

Abstract:
Robots may interact with their environment to achieve a goal in a diverse set of ways. Mechanical contact represents one of the most interesting interaction forms. Improving controllers that manage interaction may increase productivity and task quality. Control improvement through optimization techniques, however, are model-based, i.e. they require a model of the robot and/or a model of the environment, which are usually nonlinear due to contact and friction. Alternative approaches adopt evolutionary algorithms, which do not require models of the robot or the environment. Nevertheless, currently available techniques are not tested under multiple degrees of freedom conditions, or they do not take into account transition from free movement to contact condition or they take longer to converge. This paper proposes an evolutionary approach based on a genetic algorithm to optimize an impedance controller with multiple degrees of freedom. To reduce the number of experiments needed, we reduce the size of the problem based on the task’s natural constraints. The algorithm is used to perform the peg-in-hole task, and we show that convergence was achieved by the end of the evaluated generations.