Quadcopter-style drones are versatile, maneuverable, and affordable. But there is a reason you don’t hop on a quadrotor aircraft — or even a helicopter — when you go to the airport to fly across the country, and that’s because they’re inefficient. An airplane’s wings naturally provide lift as it moves forward, but a drone’s propellers need to fight gravity and provide propulsion. To gain the maneuverability of a quadcopter and the efficiency of a plane, a hybrid aircraft is ideal. But those are hard to design, which is why researchers from MIT’s Computer Science and Artificial Intelligence Laboratory have created an artificial intelligence to help with the job.
This kind of hybrid aircraft would be capable of vertical take-off and landing (VTOL) as well as traditional flight. For take off, the rotors point upwards so it can can altitude like a helicopter. Then it can transition to efficient flight by tilting the rotors forward so it’s in an airplane configuration. The problem for drones is that it’s difficult for the flight controller to handle both flights modes, and especially for it to deal with the transition between them. It has certainly been done, but it requires a lot of work. This AI from MIT CSAIL does that work with machine learning.
This AI is designed to work with OnShape, which is popular computer-aided design (CAD) software that includes simulation and analysis tools. With MIT’s machine learning system, users can create their drone design and assign the components attributes from a data set. That design is then tested with various control settings in training simulations. Eventually, the AI will find the ideal setup and output flight controller software that can be used with the real-world drone. The researchers do say that more development is needed to compensate for aerodynamic effects, but this is still a major step forward in AI-driven design.