Integration of autonomous and pattern recognition controls in hand prostheses
von Jeremy Mouchoux
Datum der mündl. Prüfung:2021-11-19
Erschienen:2022-02-08
Betreuer:Dr. Marko Markovic
Gutachter:Prof. Dr. Florentin Wörgötter
Gutachter:Prof. Dr. Arndt Schilling
Gutachter:Prof. Dr. Marcus Baum
Gutachter:Prof. Dr. Alexander Ecker
Gutachter:Prof. Dr. David Liebetanz
Gutachter:Prof. Dr. Strahinja Dosen
Dateien
Name:Integration of autonomous and pattern recogn...pdf
Size:6.40Mb
Format:PDF
Zusammenfassung
Englisch
Thanks to the progress in the robotic and mechatronic fields, upper limb prostheses became more and more dexterous, getting closer to closing the gap between the natural limb and the medical device replacing it. However, the standard prosthesis controller spread on the market is still limited in the number of functions it can efficiently control and constitute a bottleneck in prosthesis use. Two leading solutions are currently under research to tackle this discrepancy: machine learning algorithms and autonomous controllers. Myo-controllers based on machine learning algorithms enable the user to control a high but still limited number of functions directly. On the other hand, autonomous controllers are based on a diversification of the sensor modalities to integrate the context and the user's intention in control and automatise part of the grasping process, relieving the user from the physical and potentially cognitive workload associated with it. This thesis focuses on the impact of the combination of these two solutions. Therefore, two studies investigated the gain of performance gain and the range of applicability of an association of a semi-autonomous system to a machine learning myo-controller for upper limb prostheses. This dissertation introduces first a method to preshape the prosthesis based on the prediction of the user’s intended grasping strategy. This method system supports the user in real-time by preshaping the prosthetic device's hand and wrist during a reaching phase of a prehensile action. This result is achieved by merging data from inertial measurement units, computer vision, and positions and pressure sensors to reproduce artificial proprioception, artificial exteroception, and short-term memory. The autonomous controller developed was designed to support the user in dynamic objects-crowded conditions. In a second phase, it has been assessed whether a semi-autonomous system associated with a machine learning myo-controller could improve the performance compared to the same machine learning controller alone. To assess this system, eight able-bodied and two amputee participants performed a newly developed test featuring a sequence of re-localisation tasks in a scene with multiple objects. The two amputee participants completed a standard rehabilitation test as well. The results revealed that the semi-autonomous system increased the time performance and reduced the physical effort for the total duration of the trial, and more specifically, in the preshaping phase of the task. The semi-autonomous system also reduced the need to control the prosthetic wrist manually. Eventually, the last part of this dissertation focuses on the interactions between the autonomous controller and the user. It investigates the impact, through the addition of artificial error on the autonomous system output, of different shared control modalities which define the application, according to the user’s commands, of the decisions of the autonomous controller on the prosthesis. Ten able-bodied participants performed a dual-task combining a reach-and-grasp task and an auditory reaction task. Time performance, physical and cognitive workload were recorded. This test implemented different control-sharing modalities at different levels of error. The results revealed that the shared control modalities significantly impact the task performance and the physical effort required to complete the task. The effect of the level of added errors on the three different outcomes varies between the control-sharing modalities. These results, therefore, provide valuable information to design and compare semi-autonomous upper limb prosthesis systems. In conclusion, by demonstrating the benefit of semi-autonomous systems and the investigation of their application conditions, this thesis advances the development of a new generation of prosthesis controllers that automatize parts of the actuation of the prosthesis. This research enlarges the current control bottleneck, which prevents prosthesis users from taking full advantage of their highly developed device. Therefore, it can contribute to increasing the support and the autonomy that upper limb prostheses can provide to their users in daily life.
Keywords: Upper Limb; Prosthesis; Myoelectric; Computer Vision; Semi-autonomous; Artificial Propriocepetion; Artificial Exteroception; Sensor Fusion