Control implementation of compliant composite material actuators for wearable robotic exoskeleton

In previous work [1], [2], a novel actuator is presented that merges traditional electromechanical motors and multistable composite structures. It has been shown that these structures are able to arrange themselves in multi-stable configurations corresponding to local minima of their strain enery. When this composite structure is connected with the electromechanical motor as proposed, the resulting actuator shows significant benefits in terms of safety, energy saving and control implementation using the compliant property of the overall structure, the particular shape of the strain energy landscape, and the accurately predictable non-linear behavior. Hence the proposed actuator is well-suited for many robotic applications requiring continuous assistance and robust stability. In this paper, the structure's multistability property is exploited for energy saving purpose. In order to do that, a supervised learning method named Extreme Learning Machine is introduced to approximate the elastic force applied by the structure and Gradient Descent algorithm is used to compute the local minimum points equivalent to local minima of structure's strain energy.