Adaptive Control of Arm Movement based on Cerebellar Model

Abstract:

This study is an attempt to take advantage of a cerebellar model to control a biomimetic arm. Aware that a variety of cerebellar models with different levels of details has been developed, we focused on a high-level model called MOSAIC. This model is thought to be able to describe the cerebellar functionality without getting into the details of the neural circuitry. To understand where this model exactly fits, we glanced over the biology of the cerebellum and a few alternative models. Certainly, the arm control loop is composed of other components. We reviewed those elements with emphasis on modeling for our simulation. Among these models, the arm and the muscle system received the most attention. The musculoskeletal model tested independently and by means of optimization techniques, a human-like control of arm through muscle activations achieved. We have discussed how MOSAIC can solve a control problem and what drawbacks it has. Consequently, toward making a practical use of MOSAIC model, several ideas developed and tested. In this process, we borrowed concepts and methods from the control theory. Specifically, known schemes of adaptive control of a manipulator, linearization and approximation were utilized. Our final experiment dealt with a modified/adjusted MOSAIC model to adaptively control the arm. We call this model ORF-MOSAIC (Organized by Receptive Fields MOdular Selection And Identification for Control). With as few as 16 modules, we were able to control the arm in a workspace of 30 × 30 cm. The system was able to adapt to an external field as well as handling new objects despite delays. The discussion section suggests that there are similarities between microzones in the cerebellum and the modules of this new model.