Mimicking human autonomy in industrial robotic enabled sensing

Abstract

Humans have an immediate perception of shapes and surroundings through their senses and their cognitive capabilities. This innate ability enables the manual inspection of components in manufacturing environments. Trained inspectors combine their senses and handling skills with bespoke non-destructive testing instrumentation. However, manual inspections can be slow for large and/or complex geometries and prone to human factors. Automated non-destructive testing systems have emerged in recent years, to increase data acquisition speed, part coverage and inspection reliability. These tools work well when the robotic inspection takes place in a well-structured environment and an accurate part model is available. However, precisely registering the position of a part in the robot reference system makes the inspection setup very time-consuming. Furthermore, the geometry of a part may differ from its digital model, spoiling the inspection accuracy. This work introduces a new approach that mimics the human perception capability and gives full manipulation autonomy to robotic sensing applications. We use a single robotized sensor to introduce a fully autonomous single-pass geometric and volumetric inspection of complex parts. Our approach can be used to solve some key challenges in quality assurance for Industry 4.0 and can find applicability beyond robotic non-destructive testing.