Researchers at Worcester Polytechnic Institute have developed a navigation system for small aerial robots that uses ultrasound sensors and artificial intelligence to operate in conditions that typically hinder conventional drone technologies, including fog, smoke, and darkness.
The study, led by Nitin J. Sanket and published in Science Robotics, demonstrates that a lightweight sensing approach inspired by bat echolocation can allow palm-sized drones to navigate autonomously while using limited onboard power and computational resources.
The system relies on two small ultrasound sensors combined with a deep learning model trained to interpret echo patterns. The approach mimics how bats process sound to orient themselves in complex environments. According to Sanket, the design reduces the need for heavier and more energy-intensive technologies such as lidar and radar, which can also be less effective in visually degraded conditions.
Autonomous drones typically depend on a combination of cameras, sensors, and computationally intensive algorithms to perceive their surroundings. Light-based systems can be disrupted by environmental factors such as low visibility or airborne particles, while the acoustic noise generated by drone propellers complicates the use of sound-based sensing. The WPI team addressed this by incorporating an acoustic shield to reduce interference from propeller noise, enabling clearer interpretation of ultrasound signals.
The researchers modified a quadrotor drone approximately six inches wide and weighing about one pound. The drone was tested in both indoor and outdoor environments, including obstacle courses with transparent and reflective materials, as well as settings with artificial fog and snow. Some trials were conducted in complete darkness.
Across 180 test runs, the system achieved navigation success rates ranging from 72% to 100%, depending on environmental complexity. Performance declined when the drone encountered thin objects such as metal poles or narrow branches, which produced weaker echo signals.
The research was supported by the National Science Foundation. Co-authors include Manoj Velmurugan, Phillip Brush, Colin Balfour, and Richard Przybyla of TDK InvenSense.
The team indicated that future work will focus on reducing system size and weight further to extend flight duration and improve operational efficiency. Longer flight times could enhance the use of such drones in applications such as search-and-rescue missions, where operational constraints can affect outcomes.
