The U.S. Navy is helping eliminate the need for human operators in drone swarm attacks. A study led by the U.S. Naval Postgraduate School (NPS) uses artificial intelligence to enable laser weapons to better target and destroy multiple drone attacks.
Lasers can attack targets at the speed of light, so major military powers are vigorously developing laser weapons to deal with a variety of threats - the most important of which is the existence of increasingly sophisticated drones.
However, lasers are not omnipotent, and there are many problems that need to be overcome before they can become practical weapons. First, current laser systems require human operators to have certain skills in identifying and shooting targets.
Essentially, this problem can be divided into two tasks. In the case of attacking a drone, the first task is to identify the type of drone in order to determine the weak points to attack. The second is training the laser beam at that weak spot long enough to destroy or disable the target -- a tricky challenge that's bound to get trickier as autonomous drones become faster and more agile in flight.
Human operators still have a chance of successfully dealing with a single drone, but swarms of drones are a different story. Sure, a laser can flash from one target to the next in a fraction of a second, but identifying a weak point and fixing the beam on it is another thing entirely. In actual combat situations, human operators can quickly become overwhelmed. The problem will get worse as lasers improve in handling hypersonic missiles.
The U.S. Marine Corps, Naval Surface Warfare Center Dahlgren Division, Lockheed Martin, Boeing and the Air Force Research Laboratory (AFRL) are collaborating to develop a new tracking system for counter-drone lasers that uses artificial intelligence to overcome human limitations in targeting and dealing with atmospheric distortion at long ranges that can cause laser beams to deflect from their target.
The research team used a titanium alloy 3D printed miniature model of the "Reaper" drone to train the artificial intelligence system. The model scans with infrared light and radar, simulating what a full-scale drone would look like through a telescope from different angles and distances in poor visibility conditions.
The image catalog generated two datasets of 100,000 images that were used to train the artificial intelligence system to identify the drone, confirm its angle relative to the observer, find weak points and fix the beam at that point. Meanwhile, radar inputs provide data for determining the drone's course and distance. To train the system, we set up three artificial intelligence training scenarios. The first scenario uses only synthetic data, the second scenario combines synthetic and real-world data, and the third scenario uses only real-world data.
According to the U.S. Navy, the third option works best and has the smallest error.
The next step will be field testing of radar and optical tracking on real targets, using a semi-automated system with a human operator controlling some aspects of the tracking.
"We now have the model running live in the tracking system," said Eric Montag, an imaging scientist at Dahlgren. "Sometime this year we plan to demonstrate automated aim point selection within the tracking framework for a simple proof of concept," Montag added. "We don't need to fire a laser to test the auto-aim feature. There are already a few programs -- the [High Energy Laser Expeditionary (HELEX) demonstrator] being one of them -- that are interested in this technology. We've been working with them to film from their platforms with our tracking system."
The research was published in Machine Vision and Applications.