Navigating the Wild West of Drone Airspace

Drone swarms consist of a group of drones that are controlled by either a central computer, or by onboard path planning algorithms that communicate with one another. Many different applications have benefited from this technology, in both civilian and military roles. A swarm of drones can, for example, map a large geographical area quickly, inspect large infrastructure assets, or improve crop yields through precision agriculture. And as the technology continues to develop, we can expect to see even more innovative and creative uses for them.

There are still a few kinks to work out before drone swarms can reach their full potential, however. One of the primary issues developers of these systems face is in collision avoidance. With so many different flight paths to account for, it is very challenging to keep a safe distance between all of the drones at all times.

Centralized systems, in which the flight paths of all drones are coordinated by a central server, suffer from a number of issues, like a lack of scalability and a single point of failure. For reasons such as these, much effort has been put into developing decentralized solutions that will sidestep these problems. However, decentralized schemes also have a notable problem of their own — communication latency. With drones zipping all around the sky at high speeds, even relatively small delays can be enough to cause trajectory conflicts and crashes.

A team of researchers at MIT discovered this problem the hard way as they developed a multi-agent trajectory planner they call MADER. The system was designed to formulate optimal trajectories for each drone in a swarm, and prevent collisions by communicating with the other drones in the swarm. They tested MADER out in simulated environments and the results looked great — but then they got some physical drones together and ran a few real-world experiments. Let’s just say that it did not go so well for the poor little drones.

A postmortem analysis of these trials showed that the failures stemmed from communication delays between drones. If they did not have up-to-date information on where all of the other drones were headed, they could select trajectories that would put them on a direct collision course with another vehicle. With this experience under their belts, the team set out to build a new type of decentralized system that was robust against communications delays.

The details of that new system, called Robust MADER, have recently been published. To account for communications delays, Robust MADER introduces what they call a delay check. During this time, the drone continues executing its present trajectory plan while it waits for a specified amount of time. This time allows any delayed communications to come in so that the trajectories of other drones can be compared with a new flight plan that has not yet been executed. If no conflicts are detected after the delay check, the new plan is then safe to fly.

Robust MADER could, in theory, scale to swarms with thousands of drones that operate independently of any other external system. And those flights would be collision-free, as well. Real-world experiments using this system showed it to be 100 percent successful at avoiding collisions.

While drones implementing Robust MADER do not move quite as fast as other systems (3.4 meters per second), it was the only system to have a perfect track record in avoiding crashes. This performance does come with one caveat, however — an appropriate delay check time period must be selected. If it is too short, collisions can occur. If it is too long, flight paths may not be optimal.

Moving forward, the researchers intend to run larger scale experiments to confirm that Robust MADER can scale up to larger swarm sizes.