English

VAMPO: Policy Optimization for Improving Visual Dynamics in Video Action Models

Robotics 2026-03-23 v1

Abstract

Video action models are an appealing foundation for Vision--Language--Action systems because they can learn visual dynamics from large-scale video data and transfer this knowledge to downstream robot control. Yet current diffusion-based video predictors are trained with likelihood-surrogate objectives, which encourage globally plausible predictions without explicitly optimizing the precision-critical visual dynamics needed for manipulation. This objective mismatch often leads to subtle errors in object pose, spatial relations, and contact timing that can be amplified by downstream policies. We propose VAMPO, a post-training framework that directly improves visual dynamics in video action models through policy optimization. Our key idea is to formulate multi-step denoising as a sequential decision process and optimize the denoising policy with rewards defined over expert visual dynamics in latent space. To make this optimization practical, we introduce an Euler Hybrid sampler that injects stochasticity only at the first denoising step, enabling tractable low-variance policy-gradient estimation while preserving the coherence of the remaining denoising trajectory. We further combine this design with GRPO and a verifiable non-adversarial reward. Across diverse simulated and real-world manipulation tasks, VAMPO improves task-relevant visual dynamics, leading to better downstream action generation and stronger generalization. The homepage is https://vampo-robot.github.io/VAMPO/.

Keywords

Cite

@article{arxiv.2603.19370,
  title  = {VAMPO: Policy Optimization for Improving Visual Dynamics in Video Action Models},
  author = {Zirui Ge and Pengxiang Ding and Baohua Yin and Qishen Wang and Zhiyong Xie and Yemin Wang and Jinbo Wang and Hengtao Li and Runze Suo and Wenxuan Song and Han Zhao and Shangke Lyu and Zhaoxin Fan and Haoang Li and Ran Cheng and Cheng Chi and Huibin Ge and Yaozhi Luo and Donglin Wang},
  journal= {arXiv preprint arXiv:2603.19370},
  year   = {2026}
}
R2 v1 2026-07-01T11:28:52.972Z