Semantic Segmentation for Flexible and Autonomous Manufacturing

Customized mass production of boats and other vehicles requires highly complex manufacturing processes that involve a high amount of automation. Key elements to enhance the efficiency of such systems are represented by vision and sensing, which provide robots with detailed information about the working environment. In this paper, we focus on the sanding process of boat molding tools by means of a robot, proposing the use of semantic segmentation to detect the key elements involved in production and increase the automation of the production process. We demonstrate the potential of semantic segmentation in an industrial environment which differs from the domestic scenes typically considered in the literature: it features a lower degree of variability with respect to domestic scenarios, but higher performances are required in the production environment to address challenging manufacturing operations successfully. Our segmentation algorithm has been thoroughly validated on a industrial dataset that was created on purpose, whose acquisition and annotation were speeded up thanks to our optimized pipeline.