• variable stiffness actuator
• torque control
• soft robotic
• qbMove Advanced
• Iterative Learning Control

Nowadays, to work safe with Human, robotics machines are forced inside cages, limiting the possible applications.
Soft robotics wants to achieve safe human-robot interaction and cooperation even in unstructured environments. This is done by making deformable robots with soft materials or with fixed or variable compliant mechanical parts.
The compliant behaviors of several Variable Stiffness Actuators (VSAs) are hard to model and introduce non-linear terms into the system dynamic equations. For this reason, it is necessary to develop specific identification and control techniques.
Moreover, most of the present techniques, to compensate for this lack, act with predominant feedback (FB) control components. This latter could compromise the dynamic behavior of the system.
To preserve these dynamic proprieties feedforward (FF) actions would be better. In this context, one strategy, named Iterative Learning Control (ILC) was proposed. Furthermore, this method does not require knowledge of the model.
This work proposes an ILC-based torque control strategy designed to decouple the control of the stiffness from the control of link position. The intent is to achieve simultaneous trajectory and stiffness tracking.
The dynamic model of a generic VSA, and specific agonistic-antagonistic VSA, system are analyzed. The problem is to find a control action independent from the deflection model of the stiffness mechanism. The proposed solution consists in control equilibrium torque and preset angle of the actuator. This strategy was implemented in the firmware of the device.
Moreover, a parametric identification through Least Square (LS) is presented.
Finally, several experiments were performed on systems with different degrees of freedom (DoF) to test the effectiveness of the method. All these experiments were realized using qbMove Advanced variable stiffness actuators.