Search papers, labs, and topics across Lattice.
This paper introduces PolyFlow, a novel framework for flow-based generative models that integrates polytope constraints directly into the model dynamics, eliminating the need for post-hoc corrections. By employing a discrete-time flow formulation and a projection-free architecture, PolyFlow guarantees strict adherence to arbitrary polyhedral constraints while significantly reducing computational overhead. Experimental results demonstrate that PolyFlow achieves zero constraint violations and superior generative quality, outperforming existing methods in both safety and efficiency across various planning and control tasks.
Achieving zero constraint violations without sacrificing generative quality, PolyFlow redefines the landscape of safe flow-based modeling in critical applications.
While flow-based generative models have demonstrated strong performance across a wide range of domains, deploying them in safety-critical physical systems remains challenging due to strict constraint requirements. Existing approaches typically enforce safety through post-hoc corrections, which incur substantial computational overhead and may distort the learned distribution. We propose PolyFlow, a polytope-constrained flow matching framework that embeds constraints directly into the model and flow dynamics. PolyFlow introduces a discrete-time flow formulation and a projection-free architecture, which eliminate the discretization error and guarantee strict satisfaction of arbitrary polyhedral constraints, without the need for expensive iterative solvers. Experimental results show that PolyFlow achieves zero constraint violation while maintaining high distributional fidelity across a range of planning and control tasks. Compared to state-of-the-art constrained generation baselines, PolyFlow significantly reduces inference latency and demonstrates a favorable trade-off between safety, efficiency, and generative quality. Code is available on https://github.com/MJianM/PolyFlow.
MIT CSAIL