Towards an Autonomous Vision-based Inventory Drone

Logistics is a time and cost intensive task in industry. Great expense is done in the purpose of inventory. To reduce the effort we present an inventory drone prototype developed to demonstrate the feasibility of vision-based navigation inside industrial warehouse facilities with highly repetitive visual structure. Model-based visual localization against a precomputed map of the environment is used to estimate and compensate the drift of a commercially available odometry sensor and to achieve cm-accurate navigation. Our experiments showcase the capabilities of our prototype for vision-based precise indoor navigation and autonomous inventory. All computations are performed on-board the drone.