AI Paper Atlas

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

TPAMI 2026 · Probabilistic and scientific modeling › Variational and generative sampling · approach: Flow matching and flow sampling · 12 citations (OpenAlex)

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 changedWhereMeasuredEffect
replace per-step/late projection with terminal-constrained trajectory optimization (HardFlow)pretrained flow-matching planner · Robotic manipulationSafety rate: 0.76 -> 1.00Improved
use HardFlow instead of Projection-Latepretrained flow-matching planner · Robotic manipulationTotal steps (safe trials): 67 -> 52.5Improved
use HardFlow instead of unguided samplingpretrained flow-matching planner · Robotic manipulationSafety rate: 0.06 -> 1.00Improved
add Gradient Guidance on top of Projection-AllProjection-All · Robotic manipulationSafety rate: 0.46 -> 0.40Hurt
add Gradient Guidance on top of Projection-RelaxedProjection-Relaxed · Robotic manipulationSafety rate: 0.10 -> 0.04Hurt
use HardFlow instead of the best soft-constrained baseline (Gradient Guidance)pretrained flow-matching planner · Maze2D navigationSafety rate: 0.88 -> 1.00Improved
use HardFlow instead of Gradient Guidancepretrained flow-matching planner · Maze2D navigationD4RL normalized score: 1.11 -> 1.620Improved
use HardFlow instead of projection-based methodspretrained flow-matching planner · Maze2D navigationSafety rate: 0.30 -> 1.00Improved
use HardFlow instead of Projection-All + Gradient Guidancepretrained flow-matching PDE model · Burgers PDE controlControl energy: 0.36 -> 0.28Improved
use HardFlow instead of Projection-Allpretrained flow-matching PDE model · Burgers PDE controlComputation time (s): 13.6 -> 8.3Improved
use early-relaxed projection (Projection-Relaxed) instead of all-step projectionflow-matching sampler · Burgers PDE controlR_sample: 0.00 -> 1.00Hurt
add Gradient Guidance to Projection-AllProjection-All · Burgers PDE controlControl energy: 0.53 -> 0.36Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy