Anyone Can Design a Hybrid Drone With MIT CSAIL’s AI Platform

Researchers at MIT’s Computer Science and Artificial Intelligence Lab (CSAIL) have developed an AI simulation platform that allows users to design and iterate hybrid drones in a matter of hours.

Hybrid drones offer an interesting and potentially highly capable blend of aviation approaches. A hybrid drone can switch between flying like a fixed-wing aircraft and hovering like a multi-copter, combining the ability to take off and land vertically with the energy-saving benefits of gliding.

Making drone technology more accessible

From delivery to personal transport, a whole range of exploratory efforts to combine fixed-wing and VTOL aviation methods are underway already in the industry. Most have evolved as a reaction to battery life constraints, which limit operations that require the drone to cover large distances or stay airborne for extended periods.

However, the dynamics involved when developing a drone that can seize the advantages of both vertical and horizontal flight are complex. Particularly when engineers have to develop one control system for hovering, and another for gliding horizontally like a plane. Usually, these control systems are designed manually and from scratch – a process that’s difficult, expensive and time-consuming.

Which is where MIT’s CSAIL team comes into the picture. The team has developed a platform to simplify the customization and design process for hybrid drones. The system allows users to design hybrid drones of different sizes and shapes that operate using a single controller.

“Our method allows non-experts to design a model, wait a few hours to compute its controller, and walk away with a customized, ready-to-fly drone,” says MIT CSAIL grad student Jie Xu, lead author of a new paper about the work that will be presented later this month at the SIGGRAPH conference in Los Angeles.

“The hope is that a platform like this could make more these more versatile ‘hybrid drones’ much more accessible to everyone.”

Using AI to compensate for the gaps between simulations and reality

Teaching a machine how to fly is difficult. As is developing an adaptable control system. Most methods are tested in simulation but not on real hardware, a process that often leads to unexpected hiccups.

The CSAIL team have developed AI capable of bridging that gap. The team’s neural network control system has learned from previous real-world tests and applies reinforcement learning to track the differences and compensate for them.

The system has been integrated into the CAD program OnShape. Users can choose from a range of parts and simulate flight performance.