Control of a two-DoF manipulator equipped with a pnr-variable stiffness actuator

Recently, new trends in robotics have proposed variable stiffness actuators (VSA) as an alternative to the classical actuator design, based on a rigid coupling between motors and actuated joints. These novel technologies ask for novel control and planning strategies. In the present paper we consider the problem of motion planning for a specific class of VSA. This class, named passive noise rejecting VSA (pnrVSA) is capable of rejecting noise without explicitly resorting to feedback. Passive noise rejecting actuators call for new planning and control strategies. In this paper we apply an open-loop control obtained as the solution of a stochastic optimal control problem solved with path integral (PI) approach. Specific focus is on obtaining a feedback-free numerical solution which emphasizes the peculiar characteristics of the pnrVS actuator. The proposed numerical technique is applied to control a two-DOF manipulator equipped with pnrVS actuators. It is shown that stochastic optimal control can successfully simulate an highly unstable task consisting of pushing against an unstable wall in presence of instability and noise. The proposed task is reminiscent of real tasks such as screwing and carving.