Employing IMU and ArUco Marker Based Tracking to Decode the Contact Forces Exerted by Adaptive Hands

Abstract

Adaptive, underactuated, and compliant robot hands offer a promising alternative to the fully-actuated, rigid robotic devices that are typically considered for the execution of complex tasks that require significant dexterity. The increasing popularity of adaptive hands is due to their ability to extract stable grasps even under significant object pose or other environmental uncertainties, their lightweight and affordable designs and their intuitiveness and easiness of operation. Regarding possible applications, adaptive hands have been successfully used for the execution of both robust grasping and dexterous, in-hand manipulation tasks. However, the particular class of hands also suffers from certain shortcomings and drawbacks. For example, the use of underactuation leads to a post-contact reconfiguration of the fingers that may affect the force exertion capabilities of the hands during pinch grasping. In this paper, we focus on methods to predict the contact forces exerted by adaptive hands in pinch grasps, using their postcontact reconfiguration profile. The bending profiles of the fingers are recorded using ArUco trackers and IMU sensors that are embedded on the adaptive fingers and which are used to train appropriate regression models. More precisely, we examine the efficiency of the machine learning technique (Random Forests) in predicting the exerted contact forces during the reconfiguration phase of an adaptive finger. The accuracy of the proposed method is experimentally validated for a wide range of conditions, involving different prepositionings of the robot finger with respect to the employed force sensor.