Advanced Control of Variable Stiffness Actuated Robots

A. Zhakatayev, M. Rubagotti, H.A. Varol, “Closed-Loop Control of Variable Stiffness Actuated Robots via Nonlinear Model Predictive Control”, IEEE Access, vol. 3, pp. 235-248, 2015.

Variable stiffness actuation (VSA) emerged as a new paradigm in the quest for creating anthropomorphic robots with higher dexterity, efficiency and safety. VSA robots are intrinsically compliant. Elastic elements are incorporated to the joints to store energy and also isolate the link from the motors. Due to the compliant elements, the dynamics of VSA robots is complex and highly nonlinear. This becomes a challenge for the control system design, since the controller needs to handle a nonlinear, highly constrained system with short sampling intervals. Due to these limitations, control methods for these robots have been usually open-loop so far. Presumably, the system performance against external disturbances and model uncertainties can be improved with closed-loop control. Please see the presentation on various designs of variable impedance actuated robots using the following link:

Design of Variable Impedance Actuators


Figure 1. Open-loop versus closed-loop control schemes for VSA robots.

At ARMS Lab, we are developing closed-loop control algorithms for VSA robots leveraging recent advances in optimization algorithms and increases in computational resources. Specifically, we generate reference trajectories by means of open-loop optimal control, and track these trajectories via nonlinear model predictive control in a closed-loop manner (see Fig. 1). We use a powerful optimization tool, Toolkit for Automatic Control and Dynamic Optimization (ACADO), to solve optimal control problem and to generate nonlinear model predictive controller (http://acado.github.io/).

We built a two degree-of-freedom robot with antagonistic variable stiffness actuation for experimental verification of our scheme. The experiments using this robot shows the advantages of our NMPC based closed-loop control scheme compared to the partially open-loop approach in the nominal, parameter variation and external disturbance cases.


Figure 1. Solidworks design (left) and constructed prototype (right) of two DOF planar robot manipulator with antagonistic actuation for the experimental assessment of the nonlinear model predictive control (NMPC) based closed-loop method.