Inverse optimal control for differentially flat systems with application to lower-limb prosthetic devices

Prosthetic control;Learning from demonstration

Aghasadeghi, Navid

关键词: Prosthetic control;    Learning from demonstration;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/78483/AGHASADEGHI-DISSERTATION-2015.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Powered prosthetic devices have shown to be capable of restoring natural gait to amputees.However, the commercialization of these devices is faced by some challenges, inparticular in prosthetic controller design. A common control framework for these devicesis called impedance control. The challenge in the application of this framework is that itrequires the choice of many controller parameters, which are chosen by clinicians throughtrial and error for each patient. In this thesis we automate the process of choosing theseparameters by learning from demonstration. To learn impedance controller parametersfor flat-ground, we adopt the method of learning from exemplar trajectories. Since wedo not at first have exemplar joint trajectories that are specific to each patient, we useinvariances in locomotion to produce them from pre-recorded observations of unimpairedhuman walking and from measurements of the patient’s height, weight, thigh length, andshank length. Experiments with two able-bodied human subjects wearing the Vanderbiltprosthetic leg with an able-bodied adaptor show that our method recovers the samelevel of performance that can be achieved by a clinician but reduces the amount of timerequired to choose controller parameters from four hours to four minutes.To extend this framework to learning controllers for stair ascent, we need a modelof locomotion that is capable of generating exemplar trajectories for any desired stairheight. Motivated by this challenge, we focus on a class of learning from demonstrationmethods called inverse optimal control. Inverse optimal control is the problem of computinga cost function with respect to which observed trajectories of a given dynamicsystem are optimal. We first present a new formulation of this problem, based on minimizingthe extent to which first-order necessary conditions of optimality are violated.This formulation leads to a computationally efficient solution as opposed to traditionalapproaches. Furthermore, we develop the theory of inverse optimal control for the casewhere the dynamic system is differentially flat. We demonstrate that the solution furthersimplifies in this case, in fact reducing to finite-dimensional linear least-squares minimization.We show how to make this solution robust to model perturbation, sampleddata, and measurement noise, as well as provide a recursive implementation for onlinelearning. Finally, we apply our new formulation of inverse optimal control to modelhuman locomotion during stair ascent. Given sparse observations of human walkers, ourmodel predicts joint angle trajectories for novel stair heights that compare well to motioncapture data. These exemplar trajectories are then used to learn prosthetic controllersfor one subject. We show the performance of the learned controllers in a stair ascentexperiment with the subject walking with the Vanderbilt prosthetic device.