Human intention estimation and goal-driven variable admittance control in manual guidance applications

In collaborative robotics, and especially in physical human-robot interaction, the prediction of the human motion intention can strongly improve the effectiveness of the synergy. This work tackles the issue of estimating the most likely final target towards which the operator is guiding the end-effector of the manipulator to actively assist him/her in accomplishing the task. A novel inference algorithm, based on Bayesian statistics, has been developed. It predicts the most likely 3D target among a predefined set and it also takes into account the possibility that the human drives the robot towards an unknown target. Then, a variable admittance control has also been conceived. It suitably adapts its parameters to establish a directional haptic feedback that helps the human accurately reach the desired goal position. The proposed strategies allow the human to precisely reach the goal even when his/her view is obstructed by the transported object. These algorithms have been validated through point to point collaborative motions with several volunteers and an ABB IRB140 robot.