Light-WAM: Efficient World Action Models with State-Fusion Action Decoding

Hugging Face Daily Papers Papers
robot-manipulation world-action-models video-backbone action-prediction lightweight-model latent-space policy-learning

Summary

Light-WAM is a lightweight world action model for efficient robot manipulation that uses a compact video backbone and downsampled latent space for future-video supervision, achieving high performance with low inference latency.

World Action Models (WAMs) extend robot policy learning by incorporating future prediction as an additional training objective, encouraging the policy to encode task-relevant temporal structure in its representations. Current WAMs often rely on large-scale generative architectures that incur high training costs and inference latency, making them difficult to deploy as efficient closed-loop policies. We propose Light-WAM, a lightweight World Action Model for efficient robot manipulation. Specifically, it is built with a compact video backbone and performs future-video supervision in a downsampled latent space, reducing the cost of video co-training while retaining its benefits for representation learning. For action prediction, Light-WAM introduces the StateFusionActionExpert, which reads adapted states from multiple backbone layers, fuses them through learned-query pooling, and directly predicts action chunks in a single forward pass. This design provides an efficient interface between video backbone representations and robot actions, avoiding the need for heavy generative action experts. Experiments demonstrate that Light-WAM maintains strong performance on LIBERO and achieves usable multi-task performance on RoboTwin 2.0, while using only 0.44B trainable parameters. It also achieves 72.03ms inference latency with 4.1GiB peak GPU memory and improved training throughput.
Original Article
View Cached Full Text

Cached at: 06/10/26, 12:10 AM

Paper page - Light-WAM: Efficient World Action Models with State-Fusion Action Decoding

Source: https://huggingface.co/papers/2606.08242

Abstract

Light-WAM is a lightweight world action model for robot manipulation that uses a compact video backbone and downsampled latent space for efficient future-video supervision, combined with a StateFusionActionExpert for direct action prediction.

World Action Models(WAMs) extendrobot policy learningby incorporatingfuture predictionas an additional training objective, encouraging the policy to encode task-relevant temporal structure in its representations. Current WAMs often rely on large-scalegenerative architecturesthat incur high training costs andinference latency, making them difficult to deploy as efficient closed-loop policies. We propose Light-WAM, a lightweight World Action Model for efficient robot manipulation. Specifically, it is built with a compactvideo backboneand performs future-video supervision in adownsampled latent space, reducing the cost of video co-training while retaining its benefits for representation learning. For action prediction, Light-WAM introduces theStateFusionActionExpert, which reads adapted states from multiple backbone layers, fuses them throughlearned-query pooling, and directly predictsaction chunksin a single forward pass. This design provides an efficient interface betweenvideo backbonerepresentations and robot actions, avoiding the need for heavy generative action experts. Experiments demonstrate that Light-WAM maintains strong performance on LIBERO and achieves usable multi-task performance on RoboTwin 2.0, while using only 0.44B trainable parameters. It also achieves 72.03msinference latencywith 4.1GiB peak GPU memory and improvedtraining throughput.

View arXiv pageView PDFGitHub25Add to collection

Get this paper in your agent:

hf papers read 2606\.08242

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper1

#### l1ziang/lightwam-checkpoints Updatedabout 10 hours ago

Datasets citing this paper1

#### l1ziang/lightwam-offline-cache Updatedabout 10 hours ago • 50

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.08242 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

AHA-WAM:Asynchronous Horizon-Adaptive World-Action Modeling with Observation-Guided Context Routing

Hugging Face Daily Papers

AHA-WAM is an asynchronous world-action model that uses dual Diffusion Transformers to decouple world prediction from action execution, achieving efficient long-horizon planning and real-time control. It achieves state-of-the-art performance on robotic manipulation tasks with up to 92.8% success on RoboTwin and 78.3% on real-world tasks, while reaching 24.17 Hz closed-loop control.

τ_0-WM: A Unified Video-Action World Model for Robotic Manipulation

Hugging Face Daily Papers

τ_0-WM is a unified video-action world model for robotic manipulation that integrates policy learning, video prediction, and action evaluation using a shared video diffusion backbone. It shows superior performance on challenging long-horizon and fine-grained tasks.