From a recent paper:
Concept of non-stop handover of a parcel to an airborne UAV. A supply station automatically loads a parcel onto the UAV and adjustment for the handover is achieved by controlling the supply station, which tracks the UAV using a high-speed vision system.
The Ishikawa Group, led by Professor Masatoshi Ishikawa, has been working with high-speed vision and manipulation systems for over a decade. You might remember them from YouTube hits like High Speed Dribbling and Rock Paper Scissors. The system you see in the video above uses a setup common to many of their projects: two high speed (1,000 frames per second) on pan-tilt mounts that can very precisely track objects (the hook on the bottom of the drone, in this case) while simultaneously calculating their position in space. One additional camera is used to track the UAV when it’s farther away from the base station, since the high speed cameras are optimized for things that are relatively close and bright. The arm is just two linear actuators with a platform to hold the payload.
Meanwhile, the UAV itself is very deliberately nothing special. The only modification required for it to pick up packages is a simple hook, and that hook could be installed on any kind of UAV you wanted, including fixed-wing drones or hybrids. And the hook also offers flexibility in terms of what it can hold—there’s no internal payload bay that you have to stuff your payload into. If it fits on the hook and is within the drone’s weight limits and aerodynamic tolerances, you’re good to go.
The proof of concept demonstration in the video shows a drone moving at 1 m/s, with the actual handoff taking place in about 0.3 second. This is obviously much faster than any other method of loading a package onto a drone that we’ve ever seen, and the researchers say that delivery simulations show substantial improvements in delivery efficiency over short distances—up to a 65 percent improvement in deliveries per day at a distance of 1 kilometer. The benefit goes down significantly at longer ranges, since the pickup is a smaller fraction of the total delivery time, but even at 4 km, simulations suggest an improvement of up to 18 percent. Next, the researchers would like to upgrade the two-axis robot arm used for the handoff to a six-axis arm, which would be able to reload itself while also doing some speed matching to help reduce the impact of the pickup.
Realistically, the immediate benefit of a system like this might not be very significant, simply because existing drone delivery services for consumers (all of which are very much in the prototype stage) have not yet reached a system saturation point where minutes or seconds really matter all that much. And having battery life measured in a few tens of minutes means that you’ll need to stop the drone to swap batteries anyway. As batteries get better, though, and as new techniques circumvent their shortcomings, package handoffs like these could find a useful place in a drone-delivery system.
Source: IEEE Spectrum