Development and Validation of a Screw Interlock Recognition Method based on Logistic Regression

Conventional robotic screw methods rely on position control and the use of a threshold value, normally a contact force value, to detect the contact between tool and screw, so the nut-runner can be placed correctly on the workpiece. In the case of a dynamic workspace, the variance of the workpiece position will likely cause the nut-runner to be positioned incorrectly and will significantly decrease the success rate of the screw task.
This paper proposes a novel method based on a logistic regression model for flexible Human-Robot-Collaboration (HRC) screw assembly operations in highly dynamic workspaces. The proposed method is part of a screwing strategy based on impedance control and developed for HRC applications. The screwing strategy consists of a spiral movement executed by the robot while approaching the workpiece and it is used as a search procedure to find the screw. The proposed method is used to detect when the tool correctly interlocks with the screw head so the robot can proceed with the screwing process. The goal is to stop the spiral movement timely when the nut-runner has correctly interlocked with the screw head to ensure a successful screw task and avoid potential damage to the nut-runner or the workpiece.
The proposed screw interlock recognition method utilizes a logistic regression model to observe the contact forces between tool and screw head. The learning model is trained using force data collected from experiments and then its feasibility is validated with further testing.