External Force Estimation Using Recurrent Neural Network for Human-Cooperative Robots

Evaluation in 6DOF Object Manipulation

Abstract

This article proposes an external force estimator, which can detect the net external force applied from the environment in such a case that human touches the robot, discriminating it from the object-manipulating force. For this purpose, an inverse dynamics model based on a recurrent neural network is utilized to calculate the object-manipulating force from the object motion. By using the proposed external force estimator, for example, a compliant motion to the net external force can be achieved to absorb the contact force in collision between human and robot. The validity of the proposed method has been confirmed by experiments in which a robot manipulated an object with 6 degrees of freedom.