Towards model predictive control of a prosthetic leg actuated by a momentum exchange device
J.M. Kreuk (TU Delft - Mechanical Engineering)
A.J.J. van den Boom – Mentor (TU Delft - Team Ton van den Boom)
H. Vallery – Mentor (TU Delft - Biomechatronics & Human-Machine Control)
More Info
expand_more
Github repository corresponding to the thesis
https://github.com/JesperKreuk/MPC_MEDOther than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Some above the knee amputees take a smaller step with their prosthetic leg. A momentum exchange device (MED) can increase the step length by exchanging angular momentum between the device and the leg. However, there are no controllers for MEDs during regular gait. The goal of this study is to build a model predictive controller (MPC) that controls an MED to help the amputees achieve a certain step length. The MPC uses a model of an unimpaired walking human to predict its movement. Two different models are evaluated, a compass-gait biped (CGB) model and a neuromusculoskeletal (NMS) model. The parameters of the CGB model are estimated with a grey-box estimation method. An NMS model of an unimpaired human is used to simulate the controllers as well. For control, both linear and nonlinear hybrid model predictive control methods are used. By evaluating the behaviour of the controllers, insights are gained in whether the chosen models and control methods are suitable to achieve the goal. It is concluded that the prediction of the CGB model is insufficient for the control algorithm used. The model is capable of approximating the swing of the leg, but it is unable to accurately predict the future states. The NMS model may be more accurate, but is at this moment in time not practical for control. The model is computationally expensive, observing all its states is difficult and the model can currently not be initiated in any desired state, which is required for control. The results indicate that the chosen combinations of models and control methods are not well suitable to achieve the goal. Control with an unimpaired human model unfolded many difficulties and control of an impaired human is even more difficult. A control method that does not rely on an accurate step length prediction might be more suitable. However, the identified CGB model and the control methods contribute to further research on the subject or other applications within the field.