AI Helping Prevent Drone Collisions

images provided by pixabay

This post is also available in: עברית (Hebrew)

With the ongoing increase in drone traffic, experts anticipate that by 2027 there will be nearly 1 million active commercial drones in the US, delivering packages, monitoring traffic, and providing emergency assistance, etc.

A team of researchers led by the Institute for Assured Autonomy built a system using artificial intelligence that would orchestrate drone traffic by replacing some human-led processes with autonomous decision-making.

Lanier Watkins who led the research explained: “We wanted to see if different approaches using AI could handle the expected scale of these operations in a safe manner, and it did. Our simulated system leverages autonomy algorithms to enhance the safety and scalability of UAS operations below 400 feet altitude.”

According to Techxplore, the research team addressed the challenge of increasing UAS traffic by evaluating the impact of autonomous algorithms in a simulated 3D airspace. The team knew from their previous research that using collision avoidance algorithms greatly reduced accidents, and they found that adding strategic deconfliction algorithms that control traffic timing to avoid collisions made things even safer and nearly eliminated airspace accidents.

The researchers have reportedly also equipped their simulator with two aspects of realism that allow the system to make autonomous decisions to prevent collisions: “Noisy sensors” that mimic the unpredictability of real-world conditions and make the system more adaptable, and a “fuzzy interference system” that calculates the risk level for each drone based on factors ranging from proximity to obstacles to adherence to the planned route.

The team plans to enhance its simulations further by including dynamic obstacles like weather and other real-world factors for a more comprehensive representation.

Watkins concluded: “This work helps researchers understand how autonomy algorithms that protect airspace can behave when faced with noise and uncertainty in 3D-simulated airspace and underscores the need to continuously monitor the results from these autonomous algorithms to ensure they have not reached potential failure states.”