No crystal ball is needed to envision a future that engineers have in mind, one in which air taxis and other flying vehicles ferry passengers between urban locations, avoiding the growing gridlock on the ground below. Companies are already prototyping and testing such hybrid electric "flying cars" that take off and land vertically but soar through the air like winged aircraft to enable efficient flight over longer distances.
Naturally, one of the key areas of concern for these aerial vehicles is safety. The aircraft must not only stay airborne but also remain in control regardless of problems that could arise during flight-anything from gusts of wind to objects flying in their path to failing propellers. Now a Caltech team has developed an onboard Machine Learning-based control method to help such aircraft detect and compensate for disturbances so they can keep on flying. The engineers describe the new method, which they call "Neural-Fly for Fault Tolerance" (NFFT), in a paper accepted for publication in the journal IEEE Robotics and Automation Letters.