Abstract

The convergence of Generative Artificial Intelligence (GenAI) and Multi-Agent Systems (MAS) ushers a shift in paradigm how computers collaborate to accomplish challenging tasks. Generative frameworks such as Large Language Models (LLMs), diffusion planners, and autoregressive policy networks are equipping autonomous agents to generate communication protocols, role allocation, and coordination plans dynamically. This shift is elevating MAS from rigid, rule-based interaction models to decentralized, self-organizing collectives with the capability to demonstrate emergent cooperation. This paper describes the evolution of the architecture, technical issues, and future directions of generative MAS in robotics and intelligent systems with emphasis on their potential applications in disaster response, logistics, and swarm robotics.