AndyVision wanders through the aisles of a store
Researchers at the Intel Science and Technology Center (Carnegie Mellon University) are working on an inventory-assistance robot called AndyVision. The robot would assist retailers by keeping track of product shelves, noting when items need to be restocked and when something has been misplaced. It would also guide both staff and customers to specific items on demand using its map of the store and knowledge of its products. According to Professor Priya Narasimhan, the robot’s computer vision system is better than using RFID tags, which have to be attached to products by hand and which have trouble with metallic shelving.
By 2020, the group hopes to transform the retail experience with robots that can not only keep track of inventory, but also fold clothing items, restock the shelves, and help you bring your bags to the car. The problem will be convincing stores to drop $20,000 ~ $30,000 (Mitsubishi’s Wakamaru, a comparable robot platform, retailed for around that) on something that may not always function as advertised.
Perhaps this team should contact ATR, where researchers have been experimenting with robot-assisted shopping for years, using better robot platforms (the Robovie R3 being the latest example) that could be modified with the requisite Kinect sensor. Then instead of a faceless bot that looks cobbled together, it would look friendly and approachable. ATR’s Ubiquitous Market project also includes mini robots throughout a store to recommend products and track shopping behavior (a potential source of valuable consumer data).
Tracking items with computer vision is a complicated task: the robot has to search for identifiers like barcodes, text, shape, size, and color. It also uses contextual references, such as the category of item in a particular aisle or shelf. Recently, researchers at CITEC (Bielefeld University) published a video showing their own Kinect-based method for sorting stacked and otherwise cluttered objects in real-time, which could be useful in this type of scenario.