Survey Examines Evolution of Multi-Agent Systems from Classical to Foundation Model-Based Architectures
An extensive academic review meticulously explores the progression of multi-agent systems (MASs) from traditional frameworks to those leveraging large foundation models (LFMs). This research offers a comparative analysis between classical MASs (CMASs) and LFM-based MASs (LMASs) across various aspects. CMASs are evaluated within a closed-loop coordination context, focusing on four key areas: perception, communication, decision-making, and control. In contrast, LMASs utilize LFMs to advance collaboration from basic state interactions to semantic reasoning, fostering greater flexibility and adaptability in varied situations. The analysis highlights differences in architecture, operational mechanisms, adaptability, and applications. Future directions for MASs are discussed, identifying challenges and research avenues. The survey, published on arXiv under identifier 2604.18133, addresses the swift evolution of artificial intelligence and its influence on multi-agent system innovation.
Key facts
- The survey reviews multi-agent systems (MASs) evolving from classical paradigms to large foundation model (LFM)-based architectures
- Comparative analysis contrasts classical MASs (CMASs) and LFM-based MASs (LMASs)
- CMASs are reviewed across four dimensions: perception, communication, decision-making, and control
- LMASs integrate LFMs to lift collaboration from low-level state exchanges to semantic-level reasoning
- The comparison examines architecture, operating mechanism, adaptability, and application
- Future perspectives summarize open challenges and potential research opportunities
- The paper was published on arXiv with identifier 2604.18133
- The research addresses rapid advancement of artificial intelligence
Entities
Institutions
- arXiv