Title: Developing a Modular Architecture for Edge-Based Embedded AI Agent Systems

Abstract: The emergence of Large Language Models (LLMs) has facilitated the development of agentic AI capable of sophisticated reasoning and tool manipulation. However, deploying such autonomous capabilities in pervasive computing settings proves difficult, primarily because embedded microcontrollers face stringent limitations regarding memory and power consumption. Current frameworks generally presuppose server-grade resources or uninterrupted network access, thereby neglecting the specific needs of deeply embedded systems. To address this gap, this study presents a modular reference architecture for Embedded Agent Systems, designed to integrate deterministic real-time control with agentic intelligence. The proposed framework features a tiered structure that separates On-Device Agents, which run heavily compressed neural networks alongside rule-based logic, ensuring low-latency enforcement and safety across diverse fleets of autonomous hardware. Instead of relying solely on empirical benchmarks, this paper examines the architectural design principles and trade-offs concerning latency, energy efficiency, and execution reliability within resource-limited contexts.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC