Grasping and Manipulation with Soft Robotic Hands

As robots get closer to humans, their ability to safely interact with unstructured environments proves to be essential, and the adoption of soft, or compliant, end-effectors becomes necessary. Robotic hands and grippers can exhibit a soft behaviour both by passive compliant elements embedded in their hardware design, and by active stiffness control strategies. The work presented in this Thesis dealt with both aspects and was aimed at finding answers to the growing need of mathematical models and shared standards in the emerging field of soft manipulation. Starting from the classical modelling tools provided by the kinematic analysis of robotic manipulators and the quasi-static analysis of grasping, a new simulation framework for intrinsically soft fingers was defined, together with new grasp quality indices suitable for underactuated and compliant hands. Then, the closure signature model was devised. Such model cut ties with the traditional models introducing a completely new way of representing robotic hands, that focused more on their capabilities, rather than their kinematic structure. The information provided by the closure signature was found to be useful for planning power grasps with soft robotic hands. Regarding active stiffness control strategies, methods that allow to regulate the grasp stiffness of fully-actuated, torque-controlled rigid hands were conceived with a bio-inspired approach and were extensively tested. Models and control paradigms presented in this Thesis are general enough to be used with different robotic setups with respect to those that were already tested. Future work will focus on both, the further application of the achieved results, and their theoretical extension to address other open questions, including environmental constraints exploitation in grasping and benchmarking of soft manipulation.