Perceptual Generalization and Context in a Network Memory Inspired Long-Term Memory for Artificial Cognition.

Area of Science:

• Robotics
• Artificial Intelligence
• Cognitive Science

Background:

• Open-ended learning cognitive architectures require robust Long-Term Memory (LTM) systems for progressive skill acquisition.
• Existing systems often lack mechanisms for continuous perceptual space generalization and context-specific categorization.
• Robotic embodied sensorimotor apparatuses necessitate autonomous perception management.

Purpose of the Study:

• To design an LTM structure for cognitive architectures that supports experience-based decision-making and skill automation.
• To address the challenge of generalizing and categorizing perceptions within continuous perceptual spaces autonomously.
• To introduce a novel knowledge representation for perceptual classes and contexts.

Main Methods:

• Developed a relational LTM storing knowledge nuggets as artificial neural networks (ANNs) within a configural associative structure.
• Introduced P-nodes for perceptual classes and C-nodes for contexts to manage generalization and categorization.
• Implemented and evaluated the proposed LTM system in a real robotic experiment.

Main Results:

• The proposed LTM structure effectively accommodates progressive acquisition of experience-based decision capabilities.
• Autonomous categorization of perceptions in continuous perceptual spaces was achieved.
• The system demonstrated successful performance in a real robotic application.

Conclusions:

• The novel LTM design enhances robotic cognitive architectures by enabling continuous learning and autonomous perception management.
• The P-node and C-node knowledge representation facilitates context-related generalization and categorization.
• The study validates the proposed approach through practical robotic implementation and evaluation.