Role of feedback and local coupling in CNNs for locomotion control of a quadruped robot

This paper focuses on the role of sensory feedback in a network of neural oscillators used as CNN cells for locomotion control of a simulated quadruped robot. The oscillators are locally connected only through load feedback: each cell receives inputs from the neighbouring legs load signal allowing a phase synchronization that is at the basis of the emergence locomotion gait. Gait transition is then controlled by the oscillation frequency and the connectivity scheme among load signals and CNN neurons, resulting in a load dependent multi template controller.