Autonomous Visual Navigation of an Indoor Environment Using a Parsimonious, Insect Inspired Familiarity Algorithm

Area of Science:

• Computational Neuroscience
• Robotics
• Animal Behavior

Background:

• Insect navigation, particularly by bees and ants, has long been studied.
• The Navigation by Scene Familiarity Hypothesis (NSFH) offers a simple model for insect navigation using stored visual scenes.
• A key assumption of NSFH is that visual environments provide unambiguous navigational information.

Purpose of the Study:

• To test if indoor environments offer sufficient visual data for navigation based on scene familiarity.
• To develop and evaluate an insect-inspired algorithm for retracing complex routes using visual scene matching.
• To analyze the impact of sensor parameters and environmental complexity on navigation performance and aliasing.

Main Methods:

• Created a detailed visual landscape of a laboratory and corridor using 2816 panoramic images.
• Developed a scene familiarity algorithm to compare current views with stored images for path retracing.
• Systematically tested navigation performance by varying sensor morphology, inspection angles, and similarity thresholds.

Main Results:

• Pixel-by-pixel image comparisons confirmed robust translational and rotational visual information in the lab environment.
• The developed algorithm successfully retraced complex training routes, even when the destination was not visible from the start.
• Navigation performance was influenced by sensor morphology, inspection angles, and similarity thresholds, with varying degrees of aliasing.

Conclusions:

• Indoor environments possess rich visual information adequate for navigation via scene familiarity.
• The NSFH model is supported by algorithmic demonstrations of successful route retracing in a complex visual landscape.
• Environmental visual richness impacts navigation accuracy and the potential for navigational aliasing.