The human hand is an incredibly complex system with a huge spectrum of functionality. The loss of a hand is a traumatic experience usually followed by psychological and rehabilitation challenges. The interaction between engineering and science has, for a long time, made efforts to restore the functionality of a lost limb. The most advanced prosthetic hands currently available are operated via myoelectric signals. Here, the human-machine interface (HMI) usually relies on electromyography (EMG) sensors placed on the surface of the stump. Signals picked up from these sensors are used to drive the prosthesis in a one-muscle-one-movement simplistic approach (e.g. flex biceps to close the hand). Despite several major and well-known challenges related to surface EMG acquisition, this simple control approach proved to be functional with basic prosthetic grippers, but failed to translate properly when modern multi-articulated prosthetic hands reached the market. For these prostheses, cumbersome contraction patterns must be learnt from the user to properly switch between the different grasps (e.g. flex wrist three times to enable pinch grasp), killing the intuitiveness of control. That is why in the last decades, researchers spent many efforts trying to relief the amputee users from the burden of a complicated HMI, moving the learning to the machine instead via artificial intelligence algorithms. Thanks to these efforts, it is now widely accepted that machine learning algorithms can indeed facilitate the user when operating prosthetic hands with more than one degree-of-freedom, and this is confirmed also by the commercial interests behind this solution.
But if the problem is solved and the solution is already accessible on the market, why 40% of the users still reject these devices? Why the dexterity and functionality of hand prostheses is still far from being comparable to that of a biologic limb? Well, for sure there are still challenges related to the acquisition of EMG from the surface of the skin, or to the total lack of sensory feedback, challenges that coming-soon implanted solutions are successfully overcoming. However, these solutions are still under clinical investigation and won’t reach the mass before a decade, and most importantly, they cannot provide a full answer to the complex problem of restoring the human hand functionality. Such articulated problem must be addressed in parallel from different directions: more intelligent hardware must be developed for the HMI as much as for the robotic prostheses. Unfortunately, the efforts spent so far on more intelligent and autonomous robotic hardware are far way to be satisfactory. Nowadays, we have sensors technology, artificial intelligence algorithms and portable processing capabilities required to considerably improve the inherent potential of a robotic hand to take independent decisions. Semi-autonomous prosthetic hands can be a game changer, ultimately converting the conventional view of a prosthetic hand from a tool to a more complex device that interacts in an intelligent fashion with the user and the surrounding environment.
In an effort to contribute to this direction, this project addresses two main scientific and technological challenges:
1) explore the autonomous selection of the hand grasp by processing data from exteroceptive sensors. Modern proximity sensors can tell us a lot about the material and shape of the target object that we intend to grasp, and such information can be used to predispose the robotic hand for such human-object interactions. Moreover, the same information could also be used to improve the safety of robot-human interactions.
2) explore the autonomous execution of the hand grasp by processing inertial data available from the stump. Patterns of hand acceleration and digits closure velocity during the reaching-to-grasp phase can be exploited to simply replicate the biology of human-object interactions. Much information about the user motor volition can be extracted from the reaching-to-grasp movement of the stump, heavily reducing the dependencies from conventional noisy EMG sensors.
