Local Homing Navigation Based on the Moment Model for Landmark Distribution and Features

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Homing navigation relies on environmental cues for agents to return to a home location.
• Existing snapshot models compare current views with a home snapshot to determine direction.
• Limitations exist in purely visual homing methods regarding accuracy and robustness.

Purpose of the Study:

• To propose a novel homing navigation method using a moment model.
• To enhance navigation accuracy by incorporating landmark distance and visual features.
• To establish a robust reference point for homing using the center of the moment.

Main Methods:

• Developed a moment model defining landmark inertia based on feature and squared distance.
• Utilized range values of landmarks and RGB color intensity as visual features.
• Calculated the homing vector from the centers of the moment at current and home locations.
• Treated all pixels in a panoramic image as landmarks.

Main Results:

• The moment model demonstrated effective homing guidance in both real and simulated environments.
• The proposed model integrating both visual features and landmark distance outperformed methods using only visual features.
• Performance was evaluated across various methods based on feature status, distance, and coordinate alignment.

Conclusions:

• The moment model offers a superior approach to homing navigation by combining visual and spatial information.
• This method provides a unique convergence point for reliable navigation.
• The findings suggest significant potential for the moment model in autonomous agent navigation systems.