Worcester Polytechnic Institute researchers have built a drone navigation system using a dual ultrasonic sonar array that operates at milliwatt-level power, published in Science Robotics 2026. The system navigates forests, smoke, and snow conditions where camera-based vision fails entirely, and where radar or lidar draw too much power for practical battery-limited flight.
The hardware solution is blunt and effective: a U-shaped acoustic shield mounted behind the sonar array blocks propeller noise that would otherwise corrupt the return signal. A neural network then separates object echoes from background clutter in real time. The drone flew actual woodland terrain, not just simulation. All code and hardware designs are open source on GitHub under the pearwpi/Saranga repository.
The paper is worth reading for the signal processing architecture alone, specifically how the team isolates usable echo data from a platform generating significant self-noise. The open-source release means this is immediately reproducible. The next question is range, reliability in dynamic environments, and whether milliwatt sensing holds up outside controlled test conditions.
[READ ORIGINAL →]