Développement d'une méthode d’entraînement à la propulsion manuelle en fauteuil roulant sur le simulateur haptique par l'utilisation de l’intelligence artificielle

Abstract

Manual wheelchair users represent a population of over 200,000 in Canada. The action of propelling the manual wheelchair (MW) results in a repetition of propulsion cycles that represent a load on the shoulder. This loading can lead to pain and injury in the upper limbs, which can deteriorate the user's physical condition. In addition, when learning to use a MW in a rehabilitation centre, it is uncommon for emphasis to be placed on learning an effective propulsion technique. In this context, propulsion training appears to be a possible solution for improving the efficiency of propulsion technique. A previous study showed that it is possible, via deterministic training on a propulsion simulator using real-time haptic feedback, to modify the propulsion of certain subjects to increase their propulsion efficiency. However, a second study showed that this increase was accompanied by an increase in shoulder loading. In the present study, the aim was to develop a non-deterministic training method on a propulsion simulator, in order to increase subjects' propulsion efficiency without increasing the physiological cost associated with propulsion. To achieve this, artificial intelligence was used in the form of reinforcement learning training to determine a level of resistance or rolling assistance to be provided to the subject in order to modify the user's propulsion technique. The push phase during the propulsion cycle was divided into ten states, and haptic feedback was provided in the form of assistance or resistance levels. The operating parameters of the reinforcement learning algorithm were estimated according to the environment in which it was deployed. A training protocol was established to test the performance of the reinforcement learning method on valid human subjects. This training aimed to improve propulsion efficiency without increasing physiological cost. Preliminary results were obtained from 4 subjects. In some subjects and for certain conditions, the training method achieved greater efficiency and lower physiological cost for the left side. Analysis of the results also showed improvements in cadence and thrust angle in some subjects under certain conditions, compared with these same parameters during initial propulsion. This training method could make it possible to personalize propulsion training for people who need to use a FRM, or to modify propulsion with the aim of making it less physiologically demanding for people who are already experienced wheelchair users.

Date	1 Dec 2023
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Rachid Aissaoui (Supervisor)