Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Pervasive Distributed Agentic Generative AI -- A State of The Art

Gianni Molinari, Fabio Ciravegna

TL;DR

<3-5 sentence high-level summary> This survey analyzes the emergence of large-language-model (LLM) based agents within pervasive computing, detailing their perception–reasoning–action architecture (profiling, memory, planning, action) and the end-to-end deployment options across cloud, edge, and mobile environments. It assesses alignment methods (SFT, DPO, FL), memory and planning strategies, evaluation frameworks, and practical applications in smart homes, health, and smart cities, while proposing the 'Agent as a Tool' paradigm for modular, secure, and efficient specialization. Key contributions include a comprehensive taxonomy of agent topologies, benchmarks, and evaluation approaches, plus a critical discussion of memory, energy, and privacy challenges unique to pervasive settings. The work emphasizes a future where numerous specialized agent-tools operate across the edge–fog–cloud continuum, enabling robust, private, and low-latency intelligent services at all times.

Abstract

The rapid advancement of intelligent agents and Large Language Models (LLMs) is reshaping the pervasive computing field. Their ability to perceive, reason, and act through natural language understanding enables autonomous problem-solving in complex pervasive environments, including the management of heterogeneous sensors, devices, and data. This survey outlines the architectural components of LLM agents (profiling, memory, planning, and action) and examines their deployment and evaluation across various scenarios. Than it reviews computational and infrastructural advancements (cloud to edge) in pervasive computing and how AI is moving in this field. It highlights state-of-the-art agent deployment strategies and applications, including local and distributed execution on resource-constrained devices. This survey identifies key challenges of these agents in pervasive computing such as architectural, energetic and privacy limitations. It finally proposes what we called "Agent as a Tool", a conceptual framework for pervasive agentic AI, emphasizing context awareness, modularity, security, efficiency and effectiveness.

Towards Pervasive Distributed Agentic Generative AI -- A State of The Art

TL;DR

<3-5 sentence high-level summary> This survey analyzes the emergence of large-language-model (LLM) based agents within pervasive computing, detailing their perception–reasoning–action architecture (profiling, memory, planning, action) and the end-to-end deployment options across cloud, edge, and mobile environments. It assesses alignment methods (SFT, DPO, FL), memory and planning strategies, evaluation frameworks, and practical applications in smart homes, health, and smart cities, while proposing the 'Agent as a Tool' paradigm for modular, secure, and efficient specialization. Key contributions include a comprehensive taxonomy of agent topologies, benchmarks, and evaluation approaches, plus a critical discussion of memory, energy, and privacy challenges unique to pervasive settings. The work emphasizes a future where numerous specialized agent-tools operate across the edge–fog–cloud continuum, enabling robust, private, and low-latency intelligent services at all times.

Abstract

The rapid advancement of intelligent agents and Large Language Models (LLMs) is reshaping the pervasive computing field. Their ability to perceive, reason, and act through natural language understanding enables autonomous problem-solving in complex pervasive environments, including the management of heterogeneous sensors, devices, and data. This survey outlines the architectural components of LLM agents (profiling, memory, planning, and action) and examines their deployment and evaluation across various scenarios. Than it reviews computational and infrastructural advancements (cloud to edge) in pervasive computing and how AI is moving in this field. It highlights state-of-the-art agent deployment strategies and applications, including local and distributed execution on resource-constrained devices. This survey identifies key challenges of these agents in pervasive computing such as architectural, energetic and privacy limitations. It finally proposes what we called "Agent as a Tool", a conceptual framework for pervasive agentic AI, emphasizing context awareness, modularity, security, efficiency and effectiveness.

Paper Structure

This paper contains 51 sections, 7 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Agent LLM Architecture
  3. Large Language Models
  4. Agent Modules
  5. Profiling
  6. Memory
  7. Planning
  8. Action
  9. Agent Topologies
  10. Agent Evaluation
  11. Subjective Evaluation
  12. Objective Evaluation
  13. Frameworks and Benchmarks
  14. LLM-Evaluators
  15. Field of Use
  16. ...and 36 more sections

Figures (4)

  • Figure 1: The LLM Agent architecture
  • Figure 2: The transformer architecture, from vaswani2017attention
  • Figure 3: An example of profiling prompt used in lu2024ai
  • Figure 4: A topology of classic tools and agent as a tool