Multi-Agent Flow Matching with Decoupled Generative Guidance
DeGG-Flow adds decoupled guidance so multi-agent flow matching meets hard shared and private constraints.
DeGG-Flow is a framework for multi-agent flow matching that steers generation toward hard constraints without each agent needing the others' simultaneous guidance inputs. Modeling generation as a control-affine dynamical system, the authors give guidance conditions for shared requirements that depend on several agents and private requirements that depend on neighbors, plus feasibility and finite-horizon convergence guarantees and a Wasserstein bound on distribution shift. Demonstrations cover multi-robot collaboration that reconfigures an environment to cross a spatial gap and multi-object scene generation with affordance constraints. In both settings, generated objects meet the hard requirements, including at team sizes not seen in training.
- DeGG-Flow adds decoupled guidance to multi-agent flow matching.
- Guidance covers shared team constraints and private neighbor-dependent constraints.
- Authors prove feasibility, finite-horizon convergence, and a Wasserstein deviation bound.
- Demos include multi-robot gap crossing and affordance-constrained scene generation.
From the original article
1 min read at arxiv.orgGenerative modeling is widely used for producing diverse objects from complex, multimodal distributions. However, its expressivity does not, in general, come with formal guarantees that the generated objects satisfy hard constraints or requirements.
By Ruoyu Lin · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.