A Mechanical Lower Limb Exoskeleton Prototype Synthetized with a Deep Learning Based Algorithm

This paper deals with kinematic synthesis of a mechanical lower limb given a specific closed path and the corresponding orientation of a human's foot. The purpose of this work is to find a linkage that simulates natural motion, particularly walking, actuated by only one motor. Path and orientation optimization is performed by using a deep learning algorithm combined with a normalization process, which makes it possible to compare and assess the different curves generated by a four-bar mechanism. The resulting regression function approximates the lengths of a four-bar linkage with a mean-square error of 0.01 for a particular shape of four-bar linkage. The difference between the estimated path and the targeted one for this shape is of 0.13, scaled for a unit length of the four-bar's ground link, and for the orientation of 0.40 radians. In the examples presented in this paper, this algorithm returns the dimensions of four-bar linkages that generate a near-walking path, while taking the orientation of the foot into account. The results obtained with real measurements lead to the concept of a mechanical lower limb.