As graceful as drone video footage can look, drones themselves can look pretty clumsy when paired with something designed to take them down. However, a team of developers at Stanford University’s Department of Aeronautics and Astronautics has designed a drone with the dodging and ducking skills of Muhammad Ali.

In their research paper, they explain how the drone’s abilities come from an application of machine learning. The drone is fitted with motion-capture sensors that send real-time data about its surrounding environment. Some rather daunting-looking algorithms then quickly work out whether a change in its surroundings is a “threat,” and if so, calculates an appropriate move to dodge it.

The drone is designed to face all kinds of obstacles, and you can see how it fares against a fencing sword in the video below.

The drone's skills are pretty impressive as it is, but the team hopes to make its reactions even sharper in the future by using a whole range of visual, laser, and ultrasonic sensors. Additionally, they reckon this technology could someday be applied to spacecraft, self-driving cars, robotics, and even for use in the military.