Bio-inspired orientation using the polarization pattern in the sky based on artificial neural networks

Area of Science:

• Bio-inspired computing
• Insect navigation
• Computational neuroscience

Background:

• Insects utilize celestial polarized light patterns for orientation.
• Understanding insect navigation can inform artificial systems.

Purpose of the Study:

• To develop a computational orientation model (POL-ANN) inspired by insect polarization vision.
• To evaluate the model's accuracy in predicting solar direction under various sky conditions.

Main Methods:

• Developed a POL-ANN using numerically generated sky polarization patterns.
• Trained and verified the network using diverse sky conditions.
• Compared POL-ANN performance against convolutional neural network (CNN) models.
• Assessed model accuracy with varying visual field sizes (locust-like vs. Drosophila-like).

Main Results:

• Trained POL-ANN models accurately predicted the sun's direction.
• The bio-inspired POL-ANN architecture demonstrated advantages over CNN models.
• Wider visual fields improved accuracy when polarization information was incomplete.

Conclusions:

• The POL-ANN model effectively mimics insect navigation using polarized light.
• The model's performance is sensitive to the visual field's scope.
• This approach offers a promising bio-inspired method for artificial orientation systems.