Learning-Based Approach to Inverse Kinematics of Wheeled Mobile Continuum Manipulators

Continuum manipulators have been the subject of intensive research due to their inherent flexibility, allowing them to have a dexterous positioning, even in restricted environments. However, while research abounds on static continuum manipulators, a few are interested in wheeled mobile continuum manipulators (WMCMs), where the workspace is extended, and additional degrees of freedom can be used to perform more complex tasks. This article proposes a learning framework for solving the inverse kinematics (IK) of the WMCM while maintaining multiple IK solutions. The configuration space of the WMCM is discretized to reduce the infinite number of IK solutions to a finite number. Besides, the WMCM’s IK problem is transformed into a single-section IK problem of the continuum manipulator by parameterizing the mobile platform’s variables and the actuating variables of the first
$(N-1)$
sections, with
$N$
being the number of sections. The parameterization is carried out by clustering the WMCM workspace using the growing neural gas algorithm. The proposed approach is validated by performing simulations and experiments on a WMCM named Robotino XT.