Project Overview

Modern multi-agent autonomous systems, particularly in defense scenarios involving heterogeneous drones (UAVs, USVs, UUVs), must operate effectively in communication-impaired and contested environments. In these scenarios, conflicts frequently arise from disagreements between agents about information and goals—a problem that standard consensus algorithms cannot solve. The Sheaf Enriched Autonomous Multi-Agent Networks (SEAMAN) project will address this critical challenge by developing a groundbreaking mathematical framework for robust task distribution and conflict resolution. The core of this approach is a tightly-integrated framework utilizing enriched categories to create a new mathematical syntax for tasks and preferences, and cellular sheaves to model task compatibility and ensure local-to-global consistency across the agent network. This will enable decentralized and asynchronous coordination even with intermittent or sparse communication. The project will culminate in a proof-of-concept demonstration, validating the developed algorithms on unmanned ground vehicles (UGVs) and benchmarking their performance against state-of-the-art multi-agent reinforcement learning (MARL) techniques.

Sheaf Reveal

January 28, 2026

This milestone successfully translated multi-agent task agreement into a cellular sheaf framework where agent specifications are modeled as assume-guarantee contracts. This mathematical structure uses contract embeddings and Galois connections to allow heterogeneous robots with different local maps to compare and align their mission goals. The milestone demonstrated this capability through a Mine Countermeasure Mission, where the sheaf identifies logical inconsistencies—such as conflicting safety assumptions—between vehicles in a shared environment.

Slides

Funding Info