Adaptive Central Pattern Generator to Control a Modular Lower Limb Rehabilitation Exoskeleton

One of the main challenges of current exoskeletons is the adaptation to the patients' walking capabilities. The proposal of a new device made from modular joints actuators offers new ways of adaptability. However, it also requires changes in control techniques. In this contribution, we introduce the use of a central pattern generator (CPG) based on adaptive Hopf oscillators in order to achieve the decentralised architecture for the modular exoskeleton. Hebbian learning algorithm is used to train the Adaptive CPG, using the data collected from a healthy subject's gait to mimic the human walking trajectories on the hip and knee. The analysis shows how the oscillator manages to learn the training signal and proves that once the oscillator has completed the learning process it no longer needs this signal. Furthermore, its modulation properties of amplitude and frequency are demonstrated, turning out to be suitable in the exoskeletons' control. Finally, results show how the algorithm is tested in two different configurations of the modular exoskeleton prototype, confirming its functioning.