Knowledge-Enabled Robotic Agents for Shelf Replenishment in Cluttered\n Retail Environments

Abstract

Autonomous robots in unstructured and dynamically changing retail\nenvironments have to master complex perception, knowledgeprocessing, and\nmanipulation tasks. To enable them to act competently, we propose a framework\nbased on three core components: (o) a knowledge-enabled perception system,\ncapable of combining diverse information sources to cope with occlusions and\nstacked objects with a variety of textures and shapes, (o) knowledge processing\nmethods produce strategies for tidying up supermarket racks, and (o) the\nnecessary manipulation skills in confined spaces to arrange objects in\nsemi-accessible rack shelves. We demonstrate our framework in an simulated\nenvironment as well as on a real shopping rack using a PR2 robot. Typical\nsupermarket products are detected and rearranged in the retail rack, tidying up\nwhat was found to be misplaced items.\n