Machine Learning Aided Design and Analysis of a Novel Magnetically Coupled Ball Drive

Abstract

Mobile robotic platforms navigating in unstructured and dynamic environments greatly benefit from unconstrained omnidirectional locomotion. Ground robots with spherical wheels (ball-driven robots) can enable agile omnidirectional mobility over a wide range of ground terrains. Slip occurrence at the drive and ground contact surfaces decreases actuation performance, especially during rapid vehicle acceleration and navigation on graded terrains. In this paper, the design of a new magnetically coupled ball drive that uses controllable magnetic forces to increase the transmittable actuation torque and improve traction performance is described. The design uses an internal support structure to magnetically couple the spherical wheel to the chassis enabling it to function as an omnidirectional axel. Using a model of the magnetically coupled ball drive, the slip/no-slip operational window of the new design is evaluated. A support vector classification machine is trained to classify the slip/no-slip regions and identify the relative importance scores of the feature parameters in order of their sensitivity. The classification provided insight into appropriate ranges of the critical parameters that can improve traction performance. Based on the classification of the design space, multiple design and operational points were obtained to guide the design process further. Magnetostatic simulations are then used to design space efficient magnetic arrays capable of generating coupling forces in the desired range. A prototype of the new ball drive design is developed, and the premise that the magnetically coupled ball drive can improve the slip performance is experimentally tested. The results show that it is possible to control the traction forces at both drive and ground surfaces using the magnetic coupling force and substantially increase the slip performance of the ball drive using the new design.