Drones avoid obstacles thanks to motion planning algorithms

As drones become more and more prevalent and useful, there is a need to make drones smarter and to allow them to operate in harsher environments. To successfully operate in a forest environment or other very crowded environment drones need to have both obstacle detection and motion planning capabilities.

Those two things also happen to be two of the most difficult tasks for computer scientists to pull off. MIT's Computer Science and Artificial Intelligence Labs have demonstrated software that allows drones to stop with accuracy and make very tight maneuvers over, under, and around 26 different types of obstacles in a simulated forest.

The simulated forest in the video here appears to be rope and PVC pipes. The video shows a very small drone that weighs a bit more than an ounce and is only 3.5-inches rotor to rotor flying through the obstacles at up to 1 meter per second velocities.

The MIT scientists focused on having the drones recognize free spaces rather than the obstacles. The second team used a fixed wing plane that can avoid obstacles without advanced knowledge of the space whole contending with wind and other dynamics. The team used a pre-programmed library of "funnels" representing worst case behavior that was calculated with a rigorous verification algorithm.