Enhancing Shared Control via Contact Force Classification in Human-Robot Cooperative Task Execution

Abstract

In this paper, we present a novel method to support physical human-robot interaction during the execution of collaborative manipulation tasks. In the proposed approach, the robot is able to infer the operator intentions from the human contact forces, exploiting such information to properly react to the operator interventions and suitably adapt the execution of the shared task. In particular, we assume that the robotic system can autonomously generate and execute Cartesian trajectories, while a human operator can provide interventions exerting contact forces on the robot itself. The resulting robot motion is obtained by mixing in an adaptive manner the input commands provided by both the robotic control system and the human operator. In our approach, human intention estimation relies on a Neural Network capable of distinguishing the operator contact forces that support or oppose the autonomous motion planned by the robotic system. We tested the system at work in different scenarios considering simple interaction tasks performed with the 7-DOF Kuka LWR IIWA manipulator and comparing the performance obtained by a human operator with and without the assistance of the proposed system. The collected results demonstrate the effectiveness of the proposed approach.