Model predictive control for variable stiffness elasticity actuator

Modern robotics aims to bring robots out of closed environments and place them in the environment where human being exists. Robots that will work with people in the future will be an increasingly common image not only in factories but also in our homes. This goal creates a challenge for engineers and scientists. This challenge is not only to develop safe and reliable mechanical design and control systems that will provide high quality of performed task but also to create safe systems for interacting with human being. One of the possible solution for it is to integrate into structure of robot so called elastic actuators instead of classic stiff actuators. Increasing popularity of elastic actuators results from the fact that those actuators have many positive attributes in relation to classic stiff actuators. Such features are: increased shock tolerance, easier positional control in the contact issues, easier force control (which can be directly reduced to a position control problem) or high back driveability (used in tele-manipulation process or programing by hand in robotic arms). In this paper we present a position control method for Variable Stiffness Elasticity Actuator. To achieve this goal, we modelled such actuator using discrete state-space approach, first. Parameters of model used in simulations were defined with use of prototype concept of this actuator. Control method utilizes a Model Predictive Control approach to calculate input to the DC motor. Using Model Predictive Control approach we optimize changes of control variable along prediction horizon. We also present a method on how to select stiffness of the system while taking into account maximum possible values of physical spring. In our paper we present simulation of three possible types of input: step function, smooth velocity step function and sinusoidal function. Obtained results of quality of regulation for MPC are compared to PID regulator.