Research map / Probabilistic and scientific modeling / Variational and generative sampling
HardFlow: Hard-Constrained Sampling for Flow-Matching Models via Trajectory Optimization
Zeyang Li, Kaveh Alim, Navid Azizan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace per-step/late projection with terminal-constrained trajectory optimization (HardFlow) | pretrained flow-matching planner · Robotic manipulation | Safety rate: 0.76 -> 1.00 | Improved |
| use HardFlow instead of Projection-Late | pretrained flow-matching planner · Robotic manipulation | Total steps (safe trials): 67 -> 52.5 | Improved |
| use HardFlow instead of unguided sampling | pretrained flow-matching planner · Robotic manipulation | Safety rate: 0.06 -> 1.00 | Improved |
| add Gradient Guidance on top of Projection-All | Projection-All · Robotic manipulation | Safety rate: 0.46 -> 0.40 | Hurt |
| add Gradient Guidance on top of Projection-Relaxed | Projection-Relaxed · Robotic manipulation | Safety rate: 0.10 -> 0.04 | Hurt |
| use HardFlow instead of the best soft-constrained baseline (Gradient Guidance) | pretrained flow-matching planner · Maze2D navigation | Safety rate: 0.88 -> 1.00 | Improved |
| use HardFlow instead of Gradient Guidance | pretrained flow-matching planner · Maze2D navigation | D4RL normalized score: 1.11 -> 1.620 | Improved |
| use HardFlow instead of projection-based methods | pretrained flow-matching planner · Maze2D navigation | Safety rate: 0.30 -> 1.00 | Improved |
| use HardFlow instead of Projection-All + Gradient Guidance | pretrained flow-matching PDE model · Burgers PDE control | Control energy: 0.36 -> 0.28 | Improved |
| use HardFlow instead of Projection-All | pretrained flow-matching PDE model · Burgers PDE control | Computation time (s): 13.6 -> 8.3 | Improved |
| use early-relaxed projection (Projection-Relaxed) instead of all-step projection | flow-matching sampler · Burgers PDE control | R_sample: 0.00 -> 1.00 | Hurt |
| add Gradient Guidance to Projection-All | Projection-All · Burgers PDE control | Control energy: 0.53 -> 0.36 | Improved |
Closest papers by meaning
- Chance-constrained Flow Matching for High-Fidelity Constraint-aware Generation (NeurIPS 2026 · similarity 0.89)
- Optimal Flow Matching: Learning Straight Trajectories in Just One Step (NeurIPS 2024 · similarity 0.89)
- Gradient-Free Generation for Hard-Constrained Systems (ICLR 2025 · similarity 0.89)
- On the Guidance of Flow Matching (ICML 2025 · similarity 0.88)
- Physics-Constrained Flow Matching: Sampling Generative Models with Hard Constraints (NeurIPS 2025 · similarity 0.88)
- Bellman Optimal Stepsize Straightening of Flow-Matching Models (ICLR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
