A Biologically Inspired Approach for Robot Depth Estimation

Area of Science:

• Robotics and Artificial Intelligence
• Computational Neuroscience
• Computer Vision

Background:

• Autonomous service robots require integrated reasoning, perception, and action.
• Depth perception is crucial for robotic tasks, guiding navigation and interaction.
• Existing robotic systems often lack sophisticated depth cue integration.

Purpose of the Study:

• To develop and implement a novel hierarchical dorsal architecture inspired by neuroscience for robotic depth perception.
• To enhance the reasoning and decision-making capabilities of autonomous robots by leveraging accurate depth information.
• To demonstrate the robust performance of a biologically inspired approach in robotic applications.

Main Methods:

• Utilized a hierarchical four-level dorsal architecture based on neuroscience findings.
• Applied complex Gabor filters to stereo image pairs for estimating an egocentric quantitative disparity map.
• Developed a reasoning method to infer necessary data for real-time decision-making.

Main Results:

• Successfully estimated a quantitative disparity map from stereo images.
• Generated a quantitative depth scene representation serving as input for qualitative analysis.
• Experimental results demonstrated the robust performance of the biologically inspired system.

Conclusions:

• The proposed biologically inspired architecture effectively integrates depth perception for autonomous robots.
• The approach provides a robust foundation for enhancing robotic reasoning and action.
• This work highlights the potential of neuroscience-driven designs in advancing service robotics.