Landing a drone under human control onto a moving target is difficult under the best circumstances, doing it under autonomous control raises that difficulty level, and doing so under windy conditions seems near impossible. A team of engineers from MIT has designed a system that allows a quadcopter to accomplish that feat, and autonomously land on a moving vehicle under adverse conditions. Wind presents a high degree of difficulty for multi-rotor vehicles, which usually results in repeated and slow approaches.
To get around those issues, the researchers created a vision-based system with optimization-based trajectory generation to enable dynamic landings. The system uses a finite state machine (FSM), a mathematical model of computation made up of four states: standby, search, landing, and end.
Standby lets the quadrotor take off and hover above a predefined altitude above the starting point. Search provides the quadcopter with a set of simulated GPS coordinates of an uncrewed ground vehicle (UGV) and predicts a rendezvous location of where that ground vehicle will travel and flies there. When an onboard camera detects the landing platform, the FSM switches over to landing mode, which provides the drone with a direct trajectory for landing. Once landed, it switches over to end mode and stops the propellers.
The team tested the system indoors under controlled conditions, using a drone, robotic vehicle, and a set of leaf blowers to simulate windy conditions. After several unsuccessful attempts due to tracking errors, they managed to land the drone on the moving platform successfully. The engineers state they will continue to develop the system to allow for varied flight conditions, and then perform real-world tests by landing on the back of a pickup truck while outdoors.