Adaptive control of an active ankle prosthesis

The number of lower-limb amputations worldwide keeps increasing due to the ageing population and the spread of dysvascular diseases. In order to compensate for the missing limb, lower-limb amputees equipped with a prosthesis often develop asymmetric gait strategies which can lead to comorbidities. Different types of prosthetic devices exist to help transtibial lower-limb amputees recover mobility, among which active ankle prostheses are a promising technical solution as they are the only ones able to provide a net positive work during the gait cycle. Until now, the lack of efficient control strategies for active prostheses, including real-time adaptation to the users and the environment, prevents them from taking part in some daily life activities and under-exploits the full potential of these devices. This thesis is part of a long term project that consists of developing a lightweight, energy efficient and biomimetic active ankle prosthesis named ELSA. The purpose of this work is to develop an efficient control strategy able to provide adaptive assistance to the users in an attempt to recover optimal mobility. This involves compliant actuation technique and reduction of gait asymmetry throughout the walking gait cycle. The explored strategy is based on a bio-inspired control approach. It takes advantage of a cerebellum-like circuit to estimate in real-time the complex inverse dynamics model of the user and perform a compliant actuation. On the other hand, adaptable motor primitives are used to generate and easily modulate the reference joint trajectory with the idea of reducing asymmetries. The latter are monitored using IMUs and adaptive oscillators as signal learning mechanism. A preliminary validation stage is carried out using a simulation environment to show the interest of the proposed control strategy. It consists of four experiments, each of them validating a specific feature : real-time gait analysis, position tracking and learning of the inverse dynamics using LWPR, compliance analysis, and gait asymmetry correction. Also, a few experiments are performed with the latest prototype of the prosthesis for real-time gait analysis validation, and also to show the technical issues encountered. Throughout these experiments, several conclusions can be drawn. Firstly, the real-time gait analysis allows to estimate the phase variable of each leg as well as the temporal position of key gait events. Secondly, the controller is able to learn from the user dynamics and achieve very good trajectory tracking performance. Thirdly, the actuation is much more compliant than using a pure feedback regulator with equivalent tracking performance. Finally, the asymmetry correction mechanism shows mitigate results and cannot be properly validated in the considered simulation environment. To conclude, the proposed control strategy shows promising results in terms of adaptive compliant actuation learning from the user’s dynamics, and real-time gait analysis. However, the limitations induced by the simulation environment prevent from properly validating the asymmetry correction algorithm, meaning that further experiments on real users should be conducted.