The algorithm that allows drones to avoid obstacles when flying at high speed

If you like autonomous drone competitions, you will surely remember the accidents as much as the victories. In these races, teams compete to see which drones are best trained to fly at high speeds through obstacle courses. However, it happens that the higher the speed, the less stable they become, so that accidents tend to be recurrent in this type of competition.
However, the particularly fast drones are not only used to win races: they can also be used to search for survivors of a natural disaster, for example. Perhaps that is why a group of aerospace engineers from MIT have created an algorithm that helps drones find the fastest route without hitting obstacles. It is an algorithm that combines simulations of a drone flight through a virtual obstacle course along with data from experiments with real drones in physical spaces.
The researchers have verified that a drone trained with this new algorithm can fly an obstacle course twenty percent faster than a drone trained with conventional algorithms. In addition, they found it very interesting to witness some of the decisions made by the drone in question: during some races, it would take parts of the circuit with caution, or with the intention of saving energy, in order to later have greater speed and thus overtake its competitors. It seems that the algorithm makes them faster and also smarter.
“At high speeds, there are complex aerodynamic issues that are difficult to simulate, so we use real-world experiments to fill in those gaps and to discover, for example, that it may be better to slow down a little first to be faster later.” says Ezra Tal, a student in the MIT Department of Aeronautics and Astronautics. “It is this holistic approach that we use to see how we can make the trajectory as fast as possible globally.”
“These kinds of algorithms are a very valuable step forward in allowing future drones to navigate complex environments very quickly,” adds Sertac Karaman, associate professor of aeronautics and astronautics at MIT.
Training drones to fly around obstacles is relatively easy if they are going to fly slowly. This is because aerodynamic issues such as draft are not usually considered at low speeds and do not need to be taken into account to predict the behavior of the drone. But at high speeds, those effects are much more pronounced, and how vehicles manage them is much harder to predict.
The research article is published in the International Journal of Robotics Research.