Design of learning-enabled multi-agent systems

Develop architectures, specification languages, and scalable verification and control methods for learning-enabled multi-agent systems whose size, network structure, and inter-agent data dependencies make design and analysis challenging.

Background

While single-agent systems are increasingly well-understood, multi-agent scenarios introduce complex interactions and scalability issues.

The authors explicitly note the lack of clarity on how to design learning-enabled multi-agent systems and highlight the need for new specification languages and scalable methods.

References

While learning-enabled single-agent systems are fairly well understood by now, it is unclear how to design learning-enabled multi-agent systems due to their size, complex network structure, and data dependencies between agents.

— Formal Verification and Control with Conformal Prediction  (2409.00536 - Lindemann et al., 2024) in Section 7, Open Problems and Future Directions

Therefore, understanding how these structures integrate into the current collaborative learning solutions and physical network architectures is an exciting open research direction.

— From Euclidean to Graph-Structured Data: A Survey of Collaborative Learning  (2609.02984 - Bourgerie et al., 2 Sep 2026) in Section 5.5, paragraph “Open challenges of model-based techniques”

More broadly, while all architectures are in principle applicable to other production systems with OPC UA-accessible module skills, scalability to larger factory layouts and more complex module topologies remains an open question.

— LLM-Driven Multi-Agent Control for Skill-Based Smart Manufacturing  (2610.01364 - Köhle et al., 1 Oct 2026) in Section Conclusion, limitations and directions for future work

Despite the conceptual clarity this taxonomy provides, several open challenges remain. Future work should formalize the requirements identified by the taxonomy as verifiable system specifications, develop architectural components that instantiate these requirements, and empirically evaluate their effect on coordination performance in deployed systems.

— A Taxonomy on Collective Awareness  (2610.06427 - GP-Lenza et al., 5 Oct 2026) in Section Conclusions