Search papers, labs, and topics across Lattice.
This paper introduces a novel real-time, closed-loop robotic orientation control pipeline designed for precision visual inspection, addressing the challenges posed by perception noise and surface irregularities. By employing an admittance-based framework, the system effectively integrates human operator input with perception-driven surface alignment, resulting in compliant motion that adapts to real-time adjustments. Experimental validation on a 6-DOF manipulator achieved a remarkable mean orientation error of just 0.4掳, showcasing the effectiveness of the approach in industrial settings.
Real-time robotic inspection can achieve a mean orientation error of only 0.4掳 by seamlessly integrating human input with perception-driven adjustments.
Precision visual inspection underpins quality assurance across aerospace, semiconductor, and medical manufacturing, where undetected surface anomalies on high-value parts translate directly into scrap, rework, and field failures. Robotic visual inspection requires precise alignment between the end-effector and local surface geometry in the presence of perception noise and surface irregularities. In industrial settings, a human operator is often kept in the loop via teleoperation or shared autonomy, introducing real-time adjustments that render purely offline motion planning inadequate. This motivates control architectures capable of reactive, compliant behavior under combined human and perceptual uncertainty. This paper presents a novel real-time, closed-loop robotic orientation control pipeline for precision visual inspection, with an admittance-based framework that unifies operator input and perception-driven surface alignment. We design the end-effector as a virtual sphere moving through a viscous medium, such that the resulting physically interpretable mass--damper system generates synchronized, compliant motion from orientation error and operator commands. We validate the framework on a 6-DOF manipulator demonstrating stable normal-tracking and a final mean orientation error of 0.4掳.
