Hybrid Aerial and Ground-based Mobile Robot for Retail Inventory

Abstract

Radio frequency identification (RFID) technology is increasingly used in retail stores and warehouses because it can help to improve the inventory process, enable automatic checkout, and reduce shoplifting. This dissertation introduces two related robotic systems that are designed to support RFID-based inventory counting; the first is a ground-based mobile robot, and the second is an unmanned aerial vehicle (UAV). The UAV is intended to supplement the vertical reach of the ground vehicle. The ground-based robot uses two novel methods to create a map and plan its path- (1) automapping, a semiautonomous algorithm that guides the robot through the space to be inventoried, and (2) multi-layer mapping, which synthesizes a structured-light sensor (e.g. Kinect) with a light detection and ranging (LIDAR). The experimental results show that the map made by the new automapping method is as good as one made manually. The multilayer map is more accurate than the map made by traditional simultaneous location and mapping (SLAM), compared with the ground truth of the map. In this dissertation, a control algorithm for the UAV guides the UAV to a landing pad on the ground-based robot. This algorithm is named PTAM, for parallel tracking and mapping. PTAM is a camera tracking system which can work without pre-made map and landmarks. Also, it uses a 2-D camera to do a 3-D tracking. Therefore, the UAV can track its position and scale by using this algorithm. Then, it can generate a 3-D path according to inventory needs and the UAV will move according to this path point by point. The experimental result shows that the UAV can finish a pre-set flight path with tolerable error. The flight path includes taking off from a landmark, moving from point to point as pre-set, and returning to the landmark.