Investigating the Limits of Familiarity-Based Navigation

Area of Science:

• Robotics
• Computational Neuroscience
• Artificial Intelligence

Background:

• Insect-inspired navigation offers solutions for power-constrained robots.
• Familiarity-based navigation uses a single-layer neural network and Infomax learning rule.
• Previous work demonstrated navigation up to 60m.

Purpose of the Study:

• Investigate the limits of familiarity-based navigation for longer routes.
• Determine optimal parameters for effective robot navigation.
• Inform insect navigation theories and improve robotic deployments.

Main Methods:

• Challenged the method to navigate longer routes.
• Investigated performance, view acquisition rate and dimension, network size, and robustness to noise.
• Analyzed network weights and reliance on specific image areas.

Main Results:

• Effective memorization of familiar views is possible for longer routes, but decreases with reduced input view dimensions.
• Increased view acquisition rate is necessary for consistent performance with longer routes.
• Network size can be reduced with equivalent performance, offering computational and memory savings.

Conclusions:

• Familiarity-based navigation is scalable to longer routes with adjusted parameters.
• Optimal performance requires balancing view acquisition rate, input dimension, and network size.
• The method shows robustness to noise and provides insights into insect navigation mechanisms.