A missing person. Dense fog rolling in. Maybe smoke from a wildfire, or darkness after power lines fall. These are the conditions where someone needs to be found – but these are also conditions when many search and rescue drones can't fly.

Nitin Sanket sees a solution in unlikely animals: bats who navigate obstacle-filled night skies without crashing into trees, buildings or glass walls.

The Worcester Polytechnic Institute assistant professor of robotics engineering is developing tiny flying robots that use sound to navigate the same harsh conditions that ground vision-based drones.

These bat-inspired machines use ultrasonic sensors and echolocation to "see" when cameras and other optical systems go blind.

The result is the PeAR Bat, named for Sanket's Perception and Autonomous Robotics Group at WPI.

"We are building a new class of drones with new technologies built on what bats do, echolocation – which is send out a pulse of sound and listen to the weak echoes and discern where the obstacles are," Sanket said. "This is different and better from current day technology because it works in smoke, darkness, snow, which is super important for search and rescue operations.”

The technology addresses a critical gap in rescue capabilities.

Current search and rescue robots work mainly in daylight and clear conditions, but disasters don't wait for good weather.

"Search and rescue is a dull, dangerous, and dirty job that happens a lot of times in darkness because the power lines are knocked down," Sanket said.

The National Science Foundation awarded Sanket a $704,908 grant to develop the technology with undergraduate and graduate students in a campus laboratory equipped with a flying area for testing.

The technical challenges are substantial.

Propeller noise on small drones operates at the same frequency as ultrasonic sensors, creating interference up to 20 times louder than the returning echo signals.

Sanket describes it as trying to hear a friend whisper while someone shouts at you.

His team tackled the problem from multiple angles.

Hardware solutions include metamaterials that isolate and reduce noise.

Software innovations use physics-informed deep learning and artificial intelligence to filter ultrasonic signals.

The system fuses sound data with information from inertial sensors to improve reliability.

The drones also demonstrate an unexpected advantage over current technology.

Unlike Lidar systems, which use infrared light that passes through transparent materials, ultrasonic sensors detect glass walls and clear obstacles.

In demonstrations, the tiny robots autonomously stopped and backed away from plexiglass barriers, while team members noted that Lidar-equipped robots would attempt to fly straight through and crash into the barrier.

"Imagine you've seen these ads where people just go and walk into a glass wall, right? When it's perfectly clear, you don't see it. And that's exactly what current day robots are," Sanket said. "The problem is that you have cameras like our eyes or Lidars which work similar to our eyes, and it will go through this glass wall.”

The system was built using cheap, off-the-shelf technology and adapting ultrasonic sensors never designed for aerial robots.

The ultrasonic components were originally built for automatic faucets and distance detection in stationary applications.

"Our main goal is to work on the software and the algorithmic toolkit to make use of these sensors," Sanket said. "Just using the sensor itself will not give you enough data to do obstacle avoidance.”

The drones can't rely on ultrasound alone for rescue operations.

While the acoustic system handles navigation and obstacle avoidance with high power efficiency, cameras still play a role in locating survivors.

WPI graduate student Deepak Singh is developing systems that take snapshots every one to two seconds, scanning for people while the ultrasound guides flight.

Sanket, who spent more than a decade working with vision-based autonomous systems before this project, knows their limitations well.

Light's penetration power is extremely limited in fog, smoke or darkness – conditions drivers encounter regularly but that render cameras nearly useless on robots.

Sound waves don't suffer from these constraints.

That fundamental difference opens possibilities that extend far beyond search and rescue.

Sanket hopes the project will enable real-world, in-the-wild fast deployment of cheap robots in disaster zones, search and rescue, or hazardous environment monitoring with harsh conditions not possible today.

The goal, Sanket said, is to improve and reduce the cost of search and rescue by making inexpensive and energy-efficient drones that can operate where and when vision-based robots can't.

Because they're smaller, Sanket said, they're safer and more agile.

Their low cost could enable deployment in swarms, expanding the search area during rescue operations.

Other applications beyond search and rescue could include monitoring in disaster zones and hazardous environments.

The broader principles of sound-based navigation could impact fields as diverse as self-driving cars, coral reef preservation and volcano exploration.

For now, Sanket and his team continue testing in their laboratory, refining both the perception and artificial intelligence components.

They're waiting for colleagues to develop more efficient mechanical designs that can be integrated with their acoustic navigation systems.

"We think of us as perception roboticists, wherein we are focused on the sensing and the perception and the AI side of it," Sanket said. "Eventually, when colleagues come up with a better mechanical design which is more efficient, we can merge both of these things together to build a real life bat kind of robot."

The technology represents a shift in thinking about robot autonomy.

Instead of trying to make vision systems work in impossible conditions, Sanket chose a different sense altogether – one that evolution already perfected for navigating darkness.