Survivor Drones

Courtesy: UNDP Moldova

Civilian drones today are flown under very specific and safe weather and wind conditions. This is because these drones have adaptive control algorithms that are lacking. This applies to remote-controlled drones as well as autonomous ones. The direct and specific effect of various wind conditions on aircraft dynamics, performance, and stability cannot be accurately characterised as a simple mathematical model.

Engineers at Caltech have developed Neural-Fly, a deep-learning method that can help drones cope with new and unknown wind conditions in real time. Existing Machine Learning methods require a huge amount of data to train the model and adapting these large models in real-time is all but impossible. To overcome this, Neural-Fly was developed with a "separation strategy". Only a few parameters of the neural network must be updated in real time and this is achieved with a new meta-learning algorithm, which pre-trains the neural network so that only the key parameters need to be updated to effectively capture the changing environment.

After obtaining as little as 12 minutes of flying data, autonomous quadrotor drones equipped with Neural-Fly learned how to respond to strong winds so well that their performance significantly improved (as measured by their ability to precisely follow a flight path). The error rate following that flight path is around 2.5 times to 4 times smaller compared to the current state of the art drones equipped with similar adaptive control algorithms that identify and respond to aerodynamic effects but without deep neural networks.

To learn more about this research, go here.