Small-scale location identification in natural environments with deep learning based on biomimetic sonar echoes

Area of Science:

• Bioacoustics
• Echolocation
• Robotics

Background:

• Bats navigate complex environments using echolocation, emitting ultrasonic pulses and interpreting returning echoes.
• Engineered navigation systems often rely on data-intensive sensors like lidar, unlike bats' echo-based sensing.
• Vegetation clutter echoes, though unpredictable, contain environmental information.

Purpose of the Study:

• To investigate the spatial resolution of information contained within bat-like clutter echoes.
• To determine if deep learning can extract fine-grained location data from low-rate echo data.

Main Methods:

• A deep learning model (Resnet 152) was trained on echo data from a densely sampled forest environment.
• Echo data was associated with Global Positioning System (GPS) locations, which were then clustered into spatial patches.
• The neural network identified the spatial patch corresponding to sets of 1-10 echoes.

Main Results:

• The study achieved fine-grained location identification using bat-like clutter echoes.
• The spatial resolution was comparable to recreation-grade GPS accuracy under foliage.
• Effective location identification was demonstrated at very low data rates.

Conclusions:

• Bat echolocation provides a highly efficient method for environmental sensing and navigation.
• Deep learning can decode complex acoustic data for precise localization.
• This research has implications for bio-inspired navigation systems and low-data-rate localization technologies.