An Accurately Torqued Screw

Abstract

Haptic bilateral teleoperation can transmit the feeling of touch over a network. Current developments show it can be used to perform tasks over long distances by providing predictive haptic feedback to the operator paired with video feedback of the remote environment. However, predictive models fail when the remote environment behaviour is unknown. This thesis addresses unpredictable forces by focusing on remote screwdriving. Since the point at which the screw is completely tightened is unknown, predictive haptic feedback fails, and visual inspection alone cannot verify complete tightening. We propose a hybrid strategy that combines predictive and reaction-based haptic feedback to make the operator's experience of tightening a remote screw feel intuitive and controllable. Hardware experiments of the remote domain show precise angle tracking during tightening, reliable endpoint identification based on current sensing, and capture an anomaly in motor behaviour to be accounted for. A system-level simulation characterises the interaction between the operator and the remote domains and shows that the operator feels predictive feedback during tightening, is alerted at the endpoint detection, and subsequently feels reaction-based feedback. These results suggest that this hybrid strategy can provide intuitive haptic feedback for teleoperation tasks that encounter changes in torque applied by the robot. This approach lays the foundation for force-guided teleoperated tasks when models cannot predict the haptic feedback.