THE DEVELOPMENT OF VARIABLE STIFFNESS SPHERICAL ARTIFICIAL JOINTS : AN APPROACH TO INCREASE ROBOT FUNCTIONAL ANTHROPOMORPHISM

Although robots need not necessarily be anthropomorphic in order to fulfill their function, humans are an unavoidable reference when dealing with unpredictable environments. Robots designed to share an environment with humans may not be absolutely accurate. Rather, concerns of paramount importance are dexterity and safety when dealing with scenarios involving physical interaction with humans. Examples of such scenarios include, but are not limited to, emergency services carried out by firefighters and paramedics during a natural disaster, prostheses use in amputees, and heavy equipment operations carried in industrial workplaces. If we look at human dynamics, we realize that nature usually adapts their behavior to ensure self safety and avoid environment damage by suitably modulating the mechanical impedance characteristics of their bodies. Embedding in robots human features such as softness, dynamics, and wide range of motion, can pave the way to unprecedented performance in these scenarios. My research work contributes to the state of the art of anthropomorphic robotic bodies in the direction of increasing structural and control anthropomorphism. The specific problem I tackle is the design and control of variable stiffness (VSA) multi-DoFs artificial joints, which can functionally replace human body articulations of the spherical type. Examples of articulations which can be roughly described as "spherical" in our bodies are at the neck, shoulders, hips, wrists and ankles. Indeed, current robotic and prosthetic technologies are well advanced in replicating rotoidal-type articulations (elbows, knees) also with VSA mechanisms that replicate the human ability to modulate mechanical impedance. However, at the state of the art, the design and control of effective spherical joints, especially with variable stiffness, is still lagging behind. In my work, I want to contribute in this direction. At first, I propose a 3 degrees of freedom (DoFs) hybrid spherical mechanism that combines characteristics from both serial and parallel kinematic chains, in order to replicate at the best the functionality and geometry of human biological joints. The joint configuration can be modulated to optimize the design of the joint in question. The proposed idea is implemented and validated on two prototypes, a 3-DoFs robotic neck and a 3-DoFs artificial shoulder joint that can be potentially applied in both robotic and prosthetic applications. Indeed, despite the separation in their origins, it is clear that humanoids and prostheses share the common ground of trying to replicate the dynamical behavior of human beings. I leverage the proposed solution on this fact to fill some gaps in both applications. From the aspect of human motor control, implementing a variable stiffness mechanism in spherical artificial joints can add functional anthropomorphism to the structural one. Indeed, the control inputs of an agonist-antagonist type VSA (prime movers angular positions) present analogy with those of the human muscle system (the threshold length). Hence, the agonist-antagonist type VSA can introduce the ability to closely reproduce the behavior of a pair of antagonistic muscles. The analogy can be obtained from a proper tuning of the mechanical system parameters. I propose first a control strategy that can map the estimation of the muscle activations, e.g. via ElectroMyoGraphic (EMG) sensors, on a variable stiffness elbow exoskeleton, FLExo. The latter is developed for assistive purposes in strenuous daily life activities (e.g in industrial and domestic frameworks). The control policy resulting from this mapping acts, in feed- forward, so as to exploit the muscle-like dynamics of the mechanical device. Thanks to the particular structure of the actuator, the joint stiffness naturally results from that mapping. I analyze first the linear and nonlinear structural properties of an accurate neuromuscular model augmented with an external assistive torque, then I build a simulation framework (Opensim/simulink interface) to have a benchmark for testing any controller structure on a neuromusculoskeletal-assistive device system. After that, I conduct experimental tests to prove the ability of the control policy to minimize human muscle effort to zero in assistive framework and under stability and robustness guarantees. Once validated, I propose a reformulation of the control strategy for use in a proof of concept 3-DoFs VSA spherical shoulder joint whose design is based on the hybrid kinematics that I study in the first part of my work. An inertial measurement unit/EMG interface is used to map the user desired joint position and stiffness, into the artificial joint. A better version of the VSA 3-DoFs artificial spherical joint is intended to be developed in future works and integrated into a 7-DoFs anthropomorphic VSA arm. Last but not least, a series elastic actuation version of the parallel joint is tested as an add-on for industrial manipulators to approach the problem of grasping in a box (highly constrained environment) while increasing their dexterity.