Haptic-Based Threaded Insertion: Insights from Human Demonstrations

This work presents a method to utilise only haptic information in successfully completing a threaded insertion. We derive the insights for this task from human demonstrations and highlight the sufficiency of haptic data in this application, without the use of vision-based feedback or complex geometric models. We begin with human demonstrations to characterize the haptic artefacts that arise while employing backspinning motion. Force and motion data reveal a repeatable axial force transient (a spike); this signature repeats periodically for each revolution and appears for both internally and externally threaded parts of varying sizes. We then validate the same haptic cue on a robot arm. Finally, we use this insight in an end-to-end bulb insertion pipeline. A custom mechanical adapter ensures a secure grasp to enable autonomous threaded insertion of the bulb. Experimental results confirm that the force-based approach enables robust and repeatable insertion, demonstrating that haptic cues alone are sufficient for everyday threaded assemblies.