CONTROL OF MULTI-AGENT ROBOTIC SYSTEMS USING AI TECHNOLOGIES: ADVANCEMENTS IN DECENTRALIZED LEARNING AND COORDINATION

Abstract

This paper explores the transformative shift in the control of Multi-Agent Robotic Systems (MARS) driven by breakthroughs in Artificial Intelligence (AI) between 2024 and 2026. Traditional centralized control paradigms have increasingly proven inadequate for the scale and complexity of modern applications, leading to the rise of decentralized architectures. We analyze key innovations including Large Language Model (LLM)-driven algorithm discovery, Serverless Multi-Agent Reinforcement Learning (MARL), and the integration of Blockchain for secure, decentralized coordination. Specifically, we examine the "MARLess" framework and "Puzzle it Out" world models for offline learning. Our synthesis suggests that the convergence of swarm intelligence, quantum-linked automation, and multimodal path planning is enabling a new era of collaborative robotics capable of high-level reasoning and real-time adaptation in non-stationary environments.