An implementation of a muscle-like property in a variable stiffness actuator: the pnrVSA solution

In the last decades the increment of processing power of digital controllers has lead to noteworthy improvements in planning and robot control. Interesting results have been achieved in the impedance control of stiff actuators, where, thanks to very accurate force measurements and fast control loops, it has been possible to perform challenging interaction tasks. More recently, roboticists have started designing systems capable of actively varying their intrinsic compliance. From a mechanical standpoint, advances were closely tied to the development of new actuators that try to introduce at mechanical level the advantages of compliance. Series Elastic Actuators (SEA) [1] and Variable Stiffness Actuators (VSA) [2] nowadays represent an alternative to classic stiff actuators composed by electric motors and gears. Nevertheless some drawbacks are starting to emerge, because relying on active feedback in artificial agents (such as humanoid robots) might not be a practical strategy to deal with external perturbations, specifically considering the growing amount of sensors which are currently available and have to be acquired and centrally processed to perform complex actions. Humans, for instance adopt a different control strategy to cope with perturbations. Thanks to the passive mechanical properties of biological muscles, humans can reliably plan movements in presence of unpredictability and uncertainties even with significant delays in the proprioceptive (i.e. active) feedback loop. In particular, muscle co-contraction plays an important role in the active modulation of the passive musculo-skeletal compliance. During co-contraction the mechanical stiffness of the muscles is increased, which in turn increases the passive forces which resist a destabilizing motion [3]. Along this line of research and inspired by the mechanical structure of biological muscles [4], we proposed a novel design principle for actuators that possess the ability to cancel the effect of disturbances without explicitly relying on active feedback [5]. We designed a novel single-joint Variable Stiffness Actuator (VSA), based on the use of nonlinear springs and two electric motors in agonist-antagonist configuration [6] [7]. The key element of the system is a closed force path that connects the actuator output joint to the actuator frame, allowing for the implementation of a passive mechanical feedback. Thanks to the mechanical feedback the actuator possesses a unique feature, nominally the ability to augment the “passive noise rejection” (pnr). Eventually, we characterized the properties of the actuator by modeling the effect of the gear frictions on the output joint equilibrium position [8]. ACKNOWLEDGMENT This work was supported by the FP7 EU projects TACMAN (No. 610967 ICT-2013.10.1 Cognitive Systems and Robotics), and CoDyCo (No. 600716 ICT 2011.2.1 Cognitive Systems and Robotics).