The Army envisions echolocation technology’s use in aerial drones, autonomous ground vehicles, and underwater vehicles—but non-military applications abound as well.
US Army-funded research has reportedly developed an AI-driven echolocation system inspired by bats and dolphins. The new technology will enable drones and other self-guided equipment to navigate in complete darkness or other environments that tax traditional visual sensors. The new capability opens up new possibilities for battlefield and reconnaissance drones, along with many other applications, both military and non-military.
The Army Research Office and DEVCOM Ground Vehicle Systems Center sponsored research at the University of Michigan that teaches machines to interpret real-world echolocation data in a simulated environment. Sophisticated numerical solutions created virtual echoes from differently-shaped objects in a digital 3D environment. Realistic distortions enhanced the simulated echoes, which were used to train a series of “convolutional neural networks,” each programmed to detect a specific shape.
The result is a system that mimics biological echolocation, as practiced by bats and dolphins. The convolutional neural networks allow machines to identify objects by how they scatter ultrasonic pulses. This capability enables accurate navigation and detection in darkness, or low-visibility situations caused by smoke or dust.
The Army Is Training AI in a Simulated Environment
The convolutional neural networks feature a modular architecture that allows new shapes or object types to be added by training and adding new specialized networks to the system. This facilitates easy upgrades without retraining the entire system. This modularity again mimics how animals learn to recognize new obstacles, or even prey, and provides the ability for autonomous systems to adapt to dynamic or unpredictable environments.
Training the neural networks in a simulated 3D environment significantly reduces costs and development time. It also minimizes the time needed for necessary upgrades, such as adding new neural networks. In principle, this training method should also carry over to other ultrasound-based technologies, reducing costs, shortening research times, and enhancing productivity.
Some challenges are still being worked out. The most prominent of those deals with identifying low-symmetry objects, whose echoes may resemble those of other shapes. Succeeding generations will likely enhance capability by training on more diverse object shapes and data augmentation to simulate extreme conditions or environments.
AI-Powered Echolocation Could Unlock New Possibilities—Military and Civilian Alike
Reliance on sound, rather than light or electromagnetic waves, opens new avenues in which light or heat-based sensors are limited. Ultrasonic waves can penetrate environmental obstacles that hinder visual systems. The Army envisions this technology’s use in aerial drones, autonomous ground vehicles, and underwater vehicles. Non-military applications include medical imaging, search and rescue, industrial inspection, and underwater exploration.
This new capability’s emergence coincides with the Pentagon’s renewed emphasis on establishing “American Drone Dominance.” That lofty goal will not only require vast numbers of readily deployable drone systems. Top-shelf capabilities will also be necessary to overtake and exceed the already formidable lead enjoyed by Russia, China, Ukraine, and even Israel. All services are currently engaged in developing drones and other autonomous systems.
Drones equipped with echolocation sensors could prove to be a valuable asset in battle spaces that feature ever more sophisticated countermeasures. Such a drone need not be entirely disabled if it loses its optical sensors. An effective drone force will require multiple capabilities to overcome adversarial countermeasures as well as difficult or unexpected environmental challenges.
America’s new drone program emphasizes fast deployment with minimal lag time for development, procurement, and upgrades. This new program’s simulated training system fits those parameters very well, saving money and time. Army sponsorship should lead to fielding the capability sooner than later, likely across all service branches.
About the Author: William Lawson
William Lawson is a military historian focusing on World War II and 20th century conflicts and the American Civil War. His specialty is operational level warfare, especially American amphibious doctrine. He writes on history, politics, and firearms for multiple publications and historical journals. He serves on the editorial advisory board for the Saber & Scroll Journal and Military History Chronicles and is a member of the Society for Military History and the American Historical Association. Lawson is based in Virginia.
Image: Shutterstock / isoprotonic.
