A Crash Course in Avoiding Drone Crashes

MIT's navigation system helps drone swarms avoid collisions by adjusting flight paths in real-time, even with thousands in the air.

Nick Bild
10 days ago β€’ Drones
Hey! Keep your distance! (πŸ“·: S. Zhang et al.)

If you have even glanced at the news recently, then you will have almost certainly noticed that the headlines have been heavily focused on a number of aviation-related disasters. In some of these cases, at least, the tragedies occurred as a result of aircraft getting too close to one another while in flight. This is by no means a new problem β€” it is as old as aviation itself. But as the skies become ever more crowded, the challenges associated with keeping safe distances between aircraft will only grow.

In some ways, these challenges are even greater when it comes to drones than they are with conventional aircraft. Whether they are being used for a light show, infrastructure inspections, or package delivery, large numbers of drones are often intentionally programmed to fly very close to one another. Under these conditions, a slight miscalculation, or even an unexpected gust of wind, can spell disaster for the swarm.

Now, researchers at MIT have developed a potential solution to this problem. Their new artificial intelligence (AI)-powered training method ensures that large groups of autonomous drones β€” or other multi-agent robotic systems β€” can fly together safely, even in complex environments. The technique enables drones to dynamically adjust their flight paths in real-time, avoiding collisions while still achieving their primary objectives.

Ensuring safety in swarms is presently a huge challenge because traditional methods require calculating and planning the path of every single drone in relation to all the others. This approach is computationally expensive and difficult to scale. Because of these difficulties, large drone shows typically take a shortcut in which each unit follows a predetermined path, effectively closing its eyes to unexpected obstacles or changes in flight conditions. If one drone veers off course, collisions can become unavoidable.

MIT's new approach takes a different route. Instead of programming each drone with a fixed trajectory, the researchers developed a system that allows drones to continuously map their safety margins. This means each drone determines safe zones β€” focusing only on its immediate surroundings β€” and then autonomously adjusts its movements in real-time to avoid the likelihood of a collision.

This method is called Graph Control Barrier Function Plus (GCBF+), and it uses graph neural networks to model how drones interact with their environment. Rather than relying on centralized control, each drone operates independently using local information, similar to how humans navigate a crowded shopping mall by focusing only on the people nearby rather than planning a fixed path in advance.

The system first calculates an agent’s sensing radius, defining how much of its surroundings it can detect. Using this data, the AI-powered controller predicts potential conflicts and continuously updates its flight plan. The result is a dynamic, scalable safety system that works not only for a handful of drones, but even for thousands at once.

To test the system, the team conducted both real-world and simulated experiments. In one demonstration, a group of eight small quadrotor drones (Crazyflies) successfully maneuvered around each other in midair while switching positions, a task that would normally result in collisions. The drones dynamically adjusted their flight paths in real-time, staying within their computed safety zones.

In another experiment, the drones were tasked with landing on mobile robotic platforms (Turtlebots) that continuously moved in a circle. Even with unpredictable movement, the drones were able to safely navigate and land without crashing into each other.

With drone technology advancing rapidly and their use in commercial and entertainment applications growing, innovations like GCBF+ may be essential in preventing future aerial mishaps.

Nick Bild
R&D, creativity, and building the next big thing you never knew you wanted are my specialties.
Latest articles
Sponsored articles
Related articles
Latest articles
Read more
Related articles