Safe Flow Matching: Robot Motion Planning with Control Barrier Functions
🤖 AI Summary
To address insufficient safety guarantees and poor generalization in robot motion planning under unknown environments and dynamic constraints, this paper proposes a flow-matching framework integrated with Control Barrier Functions (CBFs). We introduce the novel Flow Matching Barrier Function (FMBF), which embeds formal safety verification directly into the generative process, enabling end-to-end, post-processing-free safe trajectory sampling. The method combines neural ordinary differential equation (ODE) modeling with explicit encoding of safety constraints, supporting real-time trajectory generation. Evaluated on planar navigation and 7-DoF robotic arm manipulation tasks, our approach achieves a 32% improvement in safety rate, operates 8.6× faster than diffusion-based planners, and attains over 94% generalization success on unseen obstacle configurations. These results demonstrate significant advances in both safety-critical performance and computational efficiency for autonomous robot planning.
Application Category
📝 Abstract
Recent advances in generative modeling have led to promising results in robot motion planning, particularly through diffusion and flow-based models that capture complex, multimodal trajectory distributions. However, these methods are typically trained offline and remain limited when faced with unseen environments or dynamic constraints, often lacking explicit mechanisms to ensure safety during deployment. In this work, we propose, Safe Flow Matching (SafeFM), a motion planning approach for trajectory generation that integrates flow matching with safety guarantees. By incorporating the proposed flow matching barrier functions, SafeFM ensures that generated trajectories remain within safe regions throughout the planning horizon, even in the presence of previously unseen obstacles or state-action constraints. Unlike diffusion-based approaches, our method allows for direct, efficient sampling of constraint-satisfying trajectories, making it well-suited for real-time motion planning. We evaluate SafeFM on a diverse set of tasks, including planar robot navigation and 7-DoF manipulation, demonstrating superior safety, generalization, and planning performance compared to state-of-the-art generative planners. Comprehensive resources are available on the project website: https://safeflowmatching.github.io/SafeFM/
Problem
Research questions and friction points this paper is trying to address.
Ensures robot motion safety in unseen environments
Integrates flow matching with safety guarantees
Enables real-time constraint-satisfying trajectory generation
Innovation
Methods, ideas, or system contributions that make the work stand out.
Integrates flow matching with safety guarantees
Ensures trajectories remain within safe regions
Allows direct efficient sampling of trajectories
🔎 Similar Papers
No similar papers found.
