Major delivery companies such as DHL, UPS or Amazon have been developing small drones to deliver packages. This alternative to truck delivery is expected to start operating in the near future. The advantages of it are its speed, price, safety and sustainability: parcels would not be subjected to traffic and they would be delivered within an hour, it is 10% less expensive and it means a 73% reduction in CO2 emissions when compared to truck delivery, as well as a relieve in the road traffic network. The only drawback is that the noise produced by current drones is deemed by humans as more annoying than car noise due to its high frequency. The mission of the Silent Delivery Drone project is to provide a drone delivery system that is faster, less expensive and has lower emissions than truck delivery while complying with Dutch noise regulations. The presented innovative configuration is a combination of a quadcopter, suitable for Vertically Taking-Off and Landing (VTOL) in densely populated regions, and a flying wing, optimized for the cruise phase. It consists of a horizontal propeller used during cruise and four vertical propellers for VTOL. The drone can carry a payload of up to 2.5 kg, which corresponds to 89% of the packages delivered yearly worldwide. Four packages can be delivered while flying the maximum range of 30 km. Thanks to the low required revolutions per minute, the absolute maximum noise caused by the drone is 58 dBA at take-off from 7.5 m and 25 dBA during cruise from a distance of 120 m. This meets the Dutch night noise regulations which stablish a peak noise level of 65 dBA and average noise level of 40 dBA. We believe that a fleet of Silentium drones would revolutionize the way we perceive package delivery and it would mark the next step towards a greener, smarter and more connected future.

Abstract— Stereo vision is a commonly applied method to achieve depth perception on Micro Air Vehicles (MAVs). Stereo matching algorithms are often optimized for specific environments and camera properties, using the ground truth error as a supervisor. However, in practical applications ground truth data is usually not available. Therefore, in this research, we finetune several conventional stereo matching algorithms (BM, SGBM, and ELAS) and a neural network (AnyNet) using self-supervision. The settings of the conventional algorithms are optimized with NSGA-II, using the reconstruction error and disparity density as objective functions. AnyNet is finetuned with the reconstruction error, as well as with the disparity map of conventional methods. We conclude that finetuning the parameters of conventional stereo algorithms using the reconstruction error can lead to a slight improvement in performance compared with the general settings, depending on the stereo algorithm. The performance of the conventional methods is comparable to that of AnyNet on a major portion of the image. However, removing the values with low confidence in the disparity map of ELAS and interpolating the missing disparities leads to an accuracy well above AnyNet.