Multi-objective gain optimizer for a multi-input active disturbance rejection controller: Application to series elastic actuators

Abstract

Series elastic actuators (SEA) have been gaining increasing popularity as a mechanical drive in contemporary force-controlled robotic manipulators thanks to their ability to infer the applied torque from measurements of the elastic element’s deflection. Accurate deflection control is crucial to achieve a desired output torque and, therefore, unmodelled dynamics and dynamic loads can severely compromise force fidelity. Multi-input active disturbance rejection controllers (ADRC) have the ability to estimate such disturbances affecting the plant behaviour and cancel them via an appropriate feedback controller. Thus, they offer a promising control architecture for SEA. ADRC, however, can have upwards of eight tuning parameters for each controlled state. Tuning the controller becomes quite challenging, especially in the context of multi-input, multi-objective control. This paper tackles the problem of ADRC tuning as a multi-parametric and multi-objective optimization approach. An ADRC is developed to regulate the output torque of a multi-input hybrid motor-brake–clutch SEA. The controller has a total of 22 tunable parameters. Point dominance-based nondominated sorting genetic algorithm is used to find the optimal control gains, first considering nine individual control objectives, and then in the context of multi-objective. The algorithm provides a set of potential solutions that highlight the tradeoffs between the control objectives. It is up to the discretion of the designer to select the appropriate solution that best suits a given application. The approach is validated experimentally and the results are compared with a simulated model. Experimental results confirm the suitability of the proposed approach for single and multiple control objectives in a variety of experimental scenarios and show good agreement with the analytical model.