World-Calibrated Proposal-to-Action Flow for Vision-Language-Action Models

Jie He, Wei Li, Junwen Tong, Rui Shao, Wei-Shi Zheng, Liqiang Nie · N/A · 2026

Summary

Flow-based Vision-Language-Action (VLA) policies generate action chunks by transporting samples from a task-agnostic isotropic Gaussian source.

Abstract Summary

Flow-based Vision-Language-Action (VLA) policies generate action chunks by transporting samples from a task-agnostic isotropic Gaussian source. As this source is conditioned on neither recent execution nor predicted future evolution, (i) it discards the local continuity established by recently executed motion. (ii) Even when predictive world representations are introduced, they often only condition the transport dynamics rather than determine where generation starts, how far it may deviate, or along which action directions it may expand. Building on this observation, we introduce ProAct, a world-calibrated proposal-to-action framework that makes the generative source itself predictable. (i) To preserve motion continuity, a lightweight Proposal Expert converts recent actions into a scene-aware hypothesis via one motion-anchored endpoint flow-matching step, initializing generation near the demonstrated action manifold. (ii) To jointly capture intended scene evolution and proposal-future compatibility, a prospective World Expert treats the hypothesis as a soft motion prior while predicting the task-consistent latent future. (iii) From this compatibility, the model calibrates a proposal-centered anisotropic source, where a bounded per-step extent controls the allowed deviation and a trace-normalized low-rank geometry under a condition-number budget allocates refinement over coupled translation, rotation, and gripper directions. Compared with $\pi_{0.5}$, ProAct improves performance across simulation and real-world tasks while reducing denoising steps by 50%, inference latency by up to 25.8%, and increasing throughput by up to 34.8%.

Key Points

Flow-based Vision-Language-Action (VLA) policies generate action chunks by transporting samples from a task-agnostic isotropic Gaussian source.
As this source is conditioned on neither recent execution nor predicted future evolution, (i) it discards the local continuity established by recently executed motion.
(ii) Even when predictive world representations are introduced, they often only condition the transport dynamics rather than determine where generation starts, how far it may deviate, or along which action directions it may expand.
Building on this observation, we introduce ProAct, a world-calibrated proposal-to-action framework that makes the generative source itself predictable.
(i) To preserve motion continuity, a lightweight Proposal Expert converts recent actions into a scene-aware hypothesis via one motion-anchored endpoint flow-matching step, initializing generation near the demonstrated action manifold.

ProAct

As shown in Fig. 2 , ProAct follows a propose–calibrate–refine pipeline. A Proposal Expert grounds recent motion into a scene-aware hypothesis, a prospective World Expert models latent future and proposal–future compatibility, and an Action Expert transports the calibrated source into executable actions.

This calibration determines where refinement starts, how far it may travel, and along which coupled directions it should proceed at each horizon step.

ProAct — World-Calibrated Proposal-to-Action Flow for Vision-Language-Action Models — Figure 3: World-calibrated source construction. Proposal-conditioned future features yield per-step refinement extent and low-rank geometry, defining an anisotropic source for action refinement.

Experiments

We evaluate ProAct through four questions: (Q1) Does the world-calibrated, proposal-centered source outperform strong VLA baselines? (Q2) Are gains larger when long-horizon or perturbed conditions weaken motion continuity? (Q3) Which propose–calibrate–refine components drive the gains? (Q4) Does calibration improve proposal refinement and generation efficiency?

This section addresses Q1 and Q2 on three complementary benchmarks, testing whether a calibrated source improves strong VLA baselines under saturation, distribution shift, and long-horiz...

Experiments — World-Calibrated Proposal-to-Action Flow for Vision-Language-Action Models — Figure 4: Closed-loop proposal refinement on real robots. Drawer Toy Retrieval (left) and Block Sorting (right). Enlarged frames overlay history (purple), proposal (blue), and refined action (green); orange callouts mark changes requiring departure from local motion continuity.

AI Use Disclosure

Generative AI tools were used solely to assist with English language polishing and to check formatting and presentation consistency in the manuscript. They were not used to develop research ideas, formulate hypotheses or mathematical claims, design the methodology or experiments, implement the proposed method, or analyze and interpret experimental results.

All AI-assisted edits were reviewed by the authors, who take full responsibility for the final content of the paper. The appendix is organized as follows: Sec. A : Implementation Details.

AI Use Disclosure — World-Calibrated Proposal-to-Action Flow for Vision-Language-Action Models — Figure 5: Visual attention of the Proposal Expert. Attention maps on LIBERO (top) and RoboTwin 2.0 (bottom); warmer colors highlight objects, interaction regions, and receptacles.

Sources and Demonstrations

Figures are reproduced from He et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.