A Contact Joint Sensor Using a Force-Sensor Module Embedded in a 3D Curved Surface

Human-friendly robot designs are often inspired by human joints that exhibit lightweight, dexterity, and large compressive load capacity. However, there is a significant problem when attaching sensors to contact joints inspired by human joints. Attaching traditional sensors for obtaining information on a joint is substantially complicated by the skewed rotation axes. To solve this problem, we propose a novel contact joint sensor suitable for 3D curved surfaces. The proposed contact joint sensor is composed of a contact resistance force-sensor module for obtaining distributed pressure measurements utilized to estimate the joint information via a learning method. Each force-sensor array arranged in the desired shape on a 3D curved surface measures the surface pressure transmitted through a heterogeneous force-transmit layer. The learning-based model estimates the joint angle and torque values while maintaining the estimation performance even under varying load conditions. We validated the proposed contact joint sensor with experiments involving various load conditions. The average root-mean-squared error (RMSE) values of the flexion/extension and radial/ulnar rotation angles are
$2.2\,^{\circ }$
and
$1.7\,^{\circ }$
, respectively. In addition, estimations of the torque and tension at the contact joint show good agreement with the reference values despite changes in the load conditions.