Understanding the computational models of cerebellum using robots

Abstract

Robotics has evolved since its inception and has found major applications in industries. With recent de- mand for medical and social robots, they are required to be flexible, compliant and adaptable. Today’s robots are far from performing smooth, fine co-ordinated movements that are required to be used in the fields mentioned above. Primates seem to be excellent at such tasks, especially humans who perform complex tasks in dynamic environments and the credit is attributed to the complex brain structures and bio-mechanical design. In particular, cerebellum is understood to be involved specifically in fine co-ordinated movement control. But the understanding of computations that are responsible for this functionality has seen no consensus yet. One reason for this is the number of theories that exist to ex- plain the functionality. This work focuses on reviewing existing theories and models to come up with testable model cases in a control scenario. To understand the principles behind cerebellum two differ- ent control scenarios are developed. One a control engineering approach to control the position of a DC motor and second a biological control scenario of vestibulo-ocular reflex (VOR) for image stabilization. For testing these a biologically realistic firing rate neuron model is used. For the implementation of VOR a robot head with stereo cameras is used. The purpose of VOR is to reduce image blur in the cameras. And cerebellum is known to be the adaptable block of this reflex. At the end of this work from the engi- neering control scenario it was observed that cerebellum as forward model adds stability to the system. And in the case of VOR the cerebellum was able to adapt the gains when subjected to disturbances.