Design and control of a dexterous upper-limb prosthetic system

The human hand is considered as the highest example of dexterous system capable of interacting with different objects and adapting its manipulation abilities to them. The control of poliarticulated prosthetic hands represents one important research challenge, typically aiming at replicating the manipulation capabilities of the natural hand. The success of grasping and manipulation tasks of commercial prosthetic hands is mainly related to amputee visual feedback since they are not provided either with tactile sensors or with sophisticated control. As a consequence, slippage and object falls often occur. Another important issue to consider is the preshaping of the hand, in fact it is very important to avoid injuries for the amputees giving them a prosthetic wrist. Literature studies highlight the importance of the active pronosupination and passive flexion-extension for an upper limb amputee. This thesis wants to address the specific issue of enhancing grasping and manipulation capabilities of existing prosthetic hands, by changing the control strategy and designing a new prosthetic wrist. For this purpose it proposes a multilevel control based on two distinct levels: a high and a low level. The low level control is a parallel force-position control and directly communicate with the actuators of the prosthetic hand while the high level could be a policy search learning algorithm or an actor-critic reinforcement learning (RL) combined with central pattern generators. With the RL as high level, it was carried out the control of a commercial biomechatronic hand (the IH2 hand) including the main features of recent poliarticulated prosthetic hands. The training phase of the hand was performed in simulation, the parallel force/position control was tested in simulation whereas preliminary tests were performed on the real IH2 hand. The results obtained in simulation and on the real hand provide an important evidence of the applicability of the bio-inspired neural control to real biomechatronic hand with the typical features of a hand prosthesis. With the policy search learning algorithm as high level, the control has been tested on an anthropomorphic robotic hand with prosthetic features (the IH2 hand) equipped with force sensors. Bi-digital and tri-digital grasping tasks with and without slip information have been carried out. The KUKA-LWR has been employed to perturb the grasp stability inducing controlled slip events. The acquired data demonstrate that the proposed control has the potential to adapt to changes in the environment and guarantees grasp stability, by avoiding object fall thanks to prompt slippage event detection. For what concerns the wrist, an active prono-supination and passive flexionextension modules are designed in order to avoid compensatory movements during grasps. The main intention of this thesis is to develop the control strategies and the mechanical design of the wrist previously described. A wrist/hand combined control is proposed and preliminary tests on the prono-supination module have been carried out.