A Grounded Memory System For Smart Personal Assistants

TL;DR

This paper introduces a comprehensive grounded memory system for smart personal assistants that integrates vision-language models, knowledge graphs, and retrieval-augmented generation to enhance perception, memory, and question answering capabilities.

Contribution

It presents a novel multi-component memory system combining perception, knowledge representation, and retrieval for more realistic and effective AI assistants.

Findings

Demonstrates integration of vision-language models with knowledge graphs.

Shows improved question answering through retrieval-augmented generation.

Provides a real-world example of system effectiveness.

Abstract

A wide variety of agentic AI applications - ranging from cognitive assistants for dementia patients to robotics - demand a robust memory system grounded in reality. In this paper, we propose such a memory system consisting of three components. First, we combine Vision Language Models for image captioning and entity disambiguation with Large Language Models for consistent information extraction during perception. Second, the extracted information is represented in a memory consisting of a knowledge graph enhanced by vector embeddings to efficiently manage relational information. Third, we combine semantic search and graph query generation for question answering via Retrieval Augmented Generation. We illustrate the system's working and potential using a real-world example.