Masters Thesis: Evolutionary Co-Optimisation of Control and System Parameters for a Resonating Robot Arm

This study investigates the evolutionary co-optimisation of fuzzy control and system parameters for the Resonating robot Arm (RA). The RA is a novel concept for a pick-and-place manipulator that uses a spring mechanism to reduce the required actuator torques. Since the performance of the total system depends on the combination of the spring mechanism and the controller it is difficult to find (near) optimal solutions using conventional design approaches in which the system and the controller are optimised separately. Therefore evolutionary cooptimisation is proposed in which Evolutionary Algorithms (EAs) are used to optimise the RA system as a whole. Three experiments were conducted in which the first experiment validated the use of fuzzy control and EAs to find near optimal control solutions, and the second and third experiment considered the co-optimisation of the RA with one and two degrees-of-freedom (DOF), respectively. Two types of EAs (CoSyNE and CMA-ES) and two types of fuzzy controllers (with fixed and free membership functions) were applied and their performances compared. The results revealed that evolutionary co-optimisation yields near optimal solutions for the 1-DOF RA, which require 43% less torque than the solution found through a separate optimisation of the system and control parameters. In case of the 2-DOF RA, evolutionary co-optimisation resulted in working solutions, however, no consistent convergence to near optimal solutions was found. Additionally, it was shown that for all experiments the best solutions came from the CMA-ES algorithm in combination with the fuzzy controller with free membership functions. The main conclusion drawn from this study is that evolutionary co-optimisation is an effective approach to find near optimal solutions for the 1-DOF RA, however more research is needed for it to be effectively applied to the 2-DOF RA. Master of Science Thesis S.J. Pen