Perceptual guidance of a spraying device in turbid conditions

We propose to study the problem of vision-guided robotic deposition/ablation of soft material, a class of robot tasks that are challenged by large amounts of air-suspended dust causing turbid visual conditions. We will primarily look at shotcreting, whereby one sprays concrete on a metallic mesh to assemble reinforced structures, but our work will also target neighboring problems such as bulldozing, scraping or water jetting. The goal of this thesis is to develop models that allow the robot to maintain a sense of situational awareness, and in turn use that awareness to guide manual actions that add/remove material according to pre-established requirements. We will proceed by studying the applicability of a variety of sensors (cameras, lidars, etc), and characterize their performance in different perceptual conditions. We will develop models that merge data from these sensors, and forward models of material deposition/ablation derived from robot parameters such as the throughput of a concrete spray nozzle. We will also develop robot planning and control solutions that build on situational awareness to conduct ablation/deposition action intelligently. We will demonstrate the applicability of our work on testbed hosted at the university and in the field, in collaboration with parters of the EU project that funds this thesis.