Drone Autopilot for Food Delivery
Summary Drone delivery is a popular and emerging application at present. However, existing drone delivery systems can only deliver to outdoor open spaces via GPS, and cannot directly to the interior of recipient's building. In the era of covid-19 pandemic, we aim to reduce human contact and propose a drone delivery system that can deliver packages to the doorstep or the interior of buildings, and to achieve fully automatic control of the drone by developing visual positioning technique.

To implement the UAV autopilot without GPS, we devised Vision-Based Global Positioning System (VB-GPS). It integrates results from both visual SLAM and MBL to achieve real-time, drift-free, high-precision localization.
To evaluate VB-GPS, we compared it with three state-of-the-art positioning methods – ORB-SLAM, MBL, and PoseNet – across six scenes. In our experiment, VB-GPS outperformed the other three methods, with a median error of approximately 0.27m and 0.95° on average.(ORB-SLAM: 2.00m and 1.57°; MBL: 0.62m and 1.3°,except the scene that it fails to localize; PoseNet: 9.51m and 135.71°).

Our system can provide a one-stop service from the courier company's warehouse to the customer's home directly without additional human intervention. In addition to saving manpower, we can further reduce human contact and make home quarantine more feasible. Therefore, it is believed that the proposed VB-GPS technique can be applied to real-time positioning of any first-person camera, such as self-driving cars, warehouse delivery robots, etc., and various industries with movable cameras in the future, such as: sports cameras, vehicle recorders, navigation systems, floor sweeping robots, etc.
Technical Film
Keyword food safety cosmetics and health care aquarium fish horticulture and landscape snack foods
Research Project
Research Team
More like this
Provide the latest information of AI research centers and applied industries