Obstacle avoidance in flying robots

the robot used by researchers to conduct testing.

Researchers at Harvard and MIT have been hard at work trying to understand the mechanics of in flight collision avoidance in birds. the goal of this project is to develop small Unmanned Aerial Vehicles that can navigate a forest or urban setting without crashing into trees or buildings, while still being able to fly relatively quickly.


To do this, researchers looked for inspiration from the goshawk, a bird that spends its hunting time flying between trees, yet never crashing into them.

Drones that exist today work on the principle of being able stop within the range of its sensors. this of course, presents a bit of a speed limit issue. Associate professor of aeronautics and astronautics at MIT, Emilio Frazzoli, explains that, “If I can only see up to five meters, I can only go up to a speed that allows me to stop within five meters…Which is not very fast”. If that were the case for the goshawk, it would have to fly much slower than it actually does, so how does it manage to fly so quickly without crashing?

It turns out that the Goshawk can estimate what the density of the trees are, and knows that given the tree density, it will be able to continue to find openings in trees without crashing, even though it can’t actually see the next opening.

While conducting this study, the researchers also determined that there is a speed limit for varying obstacle densities. If something flies faster than this speed limit, then they will undoubtedly crash, regardless of the information that is available to them.

The researchers set up an obstacle course, and ran tests with pigeons, observing their avoidance behavior as they flew through the obstacles. They then applies theses techniques to a robotic prototype. the results are promising, but there are inherent difficulties in maintaining control of the aircraft due to the high angles of attack necessary to preform the necessary maneuvers.

Their project video can be seen here:

Sources: