Design of a High Level Controller for Active Foot Prostheses using Gaussian Process Intent Recognition

Active prosthetic feet seem to be promising alternatives for the passive ones. These devices can potentially emulate ankle function close to that of able-bodied people. In this path, however, a number of challenges exist. One main issue is the control of such equipments. The controller here means the master controller that is in charge to provide desirable motor positions. To deal with this challenge, in this paper, we investigate on the feasibility to design a master controller based on Gaussian process (GP) regression. The aim is to develop a master controller that could be continuously used for several speeds in active foot prostheses. To this end, different input types are used to examine which scenario results in acceptable performance of the master controller and brings appropriate prediction quality. The results show that GP-based master controller has the potentials for use in active foot prostheses. The root mean square errors of the predicted and expected motor positions, were found to be between 0.8 mm and 2.1 mm for five different walking speeds.