Flowing With Purpose: Latent Action Guided Flow Matching Policies For Robotic Manipulation

2026-06-22

AI Summary

Existing flow matching methods in robotic manipulation suffer from low training efficiency and limited policy performance due to their reliance on fixed isotropic source distributions, which fail to capture the fragmented and heteroscedastic structure of action spaces. To address this, this work proposes Latent Action Guided Flow Matching (LAFM), a framework that leverages discrete motion primitives to construct an adaptive library of prior distributions in a latent action space. This library provides structure-aligned initializations for the denoising process, dynamically accommodating the heteroscedastic characteristics present in human demonstrations. By replacing the conventional single Gaussian prior with tailored base distributions, LAFM substantially simplifies trajectory generation while enhancing the expressiveness and generalization of behavior cloning. Experiments demonstrate that LAFM improves success rates by 23.4% on real-world robotic tasks and by 10.4% on the LIBERO-90 benchmark, outperforming current large-scale vision-language-action pretrained models despite using a smaller model size.

Abstract

Flow matching has recently become a new standard for behavior cloning in robotic manipulation. However, state-of-the-art flow matching policies suffer from a systematic structural mismatch: they rely on a globally fixed isotropic source distribution despite the strongly fragmented and heteroscedastic structure of robotic action spaces. This agnostic initialization forces the model to learn highly entangled vector fields, bottlenecking training efficiency and limiting overall policy performance. To address this limitation, we introduce Latent Action Guided Flow Matching (LAFM), a novel framework that replaces the monolithic Gaussian with an adaptive library of learned prior distributions. By grounding these distributions using a latent action model, LAFM maps current observations to discrete motion primitives, selecting a specialized base distribution that provides an informed, structurally aligned initialization for the denoising process. This dynamic adaptivity naturally accommodates heteroscedasticity in human demonstrations and makes transport trajectories shorter and less entangled. Empirically, LAFM substantially outperforms standard flow matching formulations, increasing task success rates by 23.4% in real-world robotic deployments and by 10.4% on the LIBERO-90 benchmark. Furthermore, we demonstrate that LAFM achieves state-of-the-art results, surpassing massively pre-trained vision-language-action models while utilizing significantly smaller architectures.

Problem

• flow matching
• robotic manipulation
• heteroscedasticity
• behavior cloning
• action space

Innovation

• Flow Matching
• Latent Action
• Heteroscedasticity
• Behavior Cloning
• Robotic Manipulation