Researchers the world over are working on drones of all sizes for all sorts of future tasks like delivering goods to areas where traffic is a big issue. Being able to deliver goods by air and having the ability to land in tight urban confines is a big deal and could help eliminate traffic and pollution in cities. Scientists are creating custom hybrid drones that promise the best of traditional multi-copter design with the efficiency and payload capacity of a fixed-wing aircraft.
The goal with hybrid drones is to create a drone that can take off and land vertically, like a multi-copter, with aspects of a normal aircraft. The challenge for designing that type of hybrid drone comes in the control difficulty. Controlling the hybrid drones often means one set of controls for vertical flight and hovering and another set of controls for forwarding flight.
The process is difficult, expensive, and time-consuming. Researchers at MIT CSAIL have made the drone customization process easier with a new system that allows users to design drones of different sizes and shapes that change from hovering to gliding using a single controller.
MIT says that its method allows non-experts to design a model, wait for a few hours to complete its controller and walk away with a ready-to-fly drone. The team uses a neural network to create adaptable control systems. One big catch right now is that the methods the team has devised have run in simulations, but not on real hardware.
MIT designed a new component in the controller for the drone that tracks differences between simulation and real-world scenarios to allow the controller to adapt its output command. The controls don’t have to store any modes and can switch from hovering to gliding and vice versa by updating the drone’s target velocity.