World Guidance: World Modeling in Condition Space for Action Generation

TL;DR

This paper introduces WoG, a framework that models future observations in a compact condition space to improve fine-grained action generation and generalization in Vision-Language-Action models, validated through extensive experiments.

Contribution

WoG proposes a novel approach to map future observations into a compact condition space, enhancing action inference and world modeling capabilities.

Findings

Outperforms existing future prediction methods in simulation and real-world tasks.

Achieves better generalization and fine-grained action generation.

Learns effectively from large-scale human manipulation videos.

Abstract

Leveraging future observation modeling to facilitate action generation presents a promising avenue for enhancing the capabilities of Vision-Language-Action (VLA) models. However, existing approaches struggle to strike a balance between maintaining efficient, predictable future representations and preserving sufficient fine-grained information to guide precise action generation. To address this limitation, we propose WoG (World Guidance), a framework that maps future observations into compact conditions by injecting them into the action inference pipeline. The VLA is then trained to simultaneously predict these compressed conditions alongside future actions, thereby achieving effective world modeling within the condition space for action inference. We demonstrate that modeling and predicting this condition space not only facilitates fine-grained action generation but also exhibits superior generalization capabilities. Moreover, it learns effectively from substantial human manipulation videos. Extensive experiments across both simulation and real-world environments validate that our method significantly outperforms existing methods based on future prediction. Project page is available at: https://selen-suyue.github.io/WoGNet/