This Bat-Inspired AI Lets Drones "See" Through Fog and Smoke

Drones, especially quadcopter drones, drain batteries very rapidly, greatly limiting their flight time. Keeping the propellers spinning takes up most of this energy, but that’s hardly the only thing slurping down power. Sensors used by the flight controller — without which the drone would be unable to navigate — also greedily gobble up energy. For this reason, developing more efficient sensing systems holds a lot of promise for keeping drones in the sky longer.

A group led by researchers at Worcester Polytechnic Institute has just published its work in which they experimented with a novel sensing system. Inspired by bats, their technology uses ultrasound in conjunction with an artificial intelligence algorithm to help drones find their way around while using very little power. As a bonus, this method also makes it possible to see through fog, smoke, and other visual obstructions.

The research focuses on enabling small, palm-sized aerial robots to operate in environments that would typically defeat conventional navigation systems. Cameras and lidar, for example, struggle in darkness or poor weather, while radar systems are often too bulky and power-hungry for lightweight drones.

To overcome these limitations, the team developed a system called “Saranga,” which relies on just two tiny ultrasound sensors. Much like bats emitting chirps and interpreting echoes, the drone sends out sound waves and analyzes the returning signals to detect obstacles. However, interpreting these weak echoes is challenging due to interference from the drone’s own propellers. To address this, the researchers added an acoustic shield to block noise and trained a deep learning model to extract meaningful patterns from noisy data.

The result is an impressively efficient sensing system that consumes only about 1.2 milliwatts of power — orders of magnitude lower than traditional approaches. This low power requirement is especially important for small drones with limited battery capacity, where every milliwatt saved can translate into longer flight times.

In testing, the team equipped a compact quadrotor drone roughly six inches across and weighing about one pound with the system. The drone successfully navigated both indoor and outdoor obstacle courses, including environments filled with fog, darkness, and artificial snow. Across 180 trials, it achieved success rates ranging from 72% to 100%, demonstrating excellent performance under challenging conditions.

However, there were some limitations. The system struggled to detect very thin objects, such as narrow metal poles or small tree branches, which reflect only weak ultrasound signals. Even so, the results mark a significant step forward in autonomous navigation for small aerial robots.

Looking ahead, the researchers aim to further miniaturize the system and improve flight speed and endurance.