Contact Sensing via Active Oscillatory Actuation

Abstract

Collaborative robotic approaches seek to incorporate either direct human intervention in tasks previously suited for isolated robot devices, or to use precise machines (cobots) to assist in sensitive tasks. Cobots work alongside humans to extend the scope of robot assistance to spaces such as service and dynamic industrial or assembly tasks. With that said, the close proximity of humans with machines necessitates safe interaction, which can be achieved via lightweight materials and novel sensing capabilities. The inherent physical strength needed in some robot tasks, such as in assembly, make contact sensing of particular concern when introducing a human collaborator. A minimally intrusive method that can be seamlessly subsumed into extant devices is desired. In this paper, such a contact sensor is prototyped and tested. The sensor is bidirectional in that it actively provides an oscillatory actuation signal to a rigid link while simultaneously recording and analyzing the mechanical vibration of said link. Natural oscillation frequency shifts and energy concentration changes due to damping are congruent with different types of contact with the rigid link. The method is lightweight, low-cost and can be quickly incorporated into various manipulators. The developed configuration is advantageous as it does not require any delicate sensors on the robot body and relies primarily on actuated oscillations of the manipulator. Oscillatory acceleration data is collected and subsequently used to train and classify different contact locations using frequency-based features. Three separate classes are distinguished according to contact location. Results are promising and show excellent classification of both contact and contact location.