Robust Autonomous Navigation in Dynamic Environments with a Modular Multi-Agent AI Architecture for Mobile Robots

Abstract

Precision and adaptability in navigation are still a pressing issue in mobile robotics, especially when dealing with dynamic obstacles, noisy sensor data, and real-time decisionmaking. This work proposes a modular, five-layer agentic AI architecture for autonomous robotic navigation that allows lowlevel sensing to be integrated with high-level reasoning. A Perception Agent, which detects visual obstacles in real time from an IP camera; Safety Agent, which has an override authority; a Decision Agent, which fuses visual and proximity data and utilizes a Large Language Model (LLM) for adaptive navigation strategies; Control Agent, which translates decisions into hardware-level commands; and Monitoring Agent, which logs state continuously and performs diagnostics. An Orchestration Agent coordinates all layers, and a live monitoring and control dashboard is provided via Flask. Experimental results verify that the proposed architecture has enhanced the reliability of navigation, improved response time in case of obstacles, and increased adaptability in decisions. It confirms the efficiency of combining agentic AI principles with real-time sensor fusion to ensure robust, scalable autonomous navigation.