Faster-WAM: Do World Action Models Need Deep Action Modules?

2026-08-03

AI Summary

Existing World Action Models (WAMs) tightly couple action modules with video backbones, resulting in high computational overhead and inference latency. This work proposes the Dock of Transformer (DoT) architecture, which centers around a pretrained video Transformer as a shared representation hub and connects lightweight task heads via modular docking interfaces, enabling flexible head design and direct access to features from all backbone layers. Introducing a novel “video-centric” paradigm, DoT employs a single-layer action head interfaced with a 30-layer backbone—achieving efficient inference and strong generalization without additional embodied pretraining. Enhanced by multi-layer key-value fusion and RoPE realignment, the lightweight action head attains state-of-the-art performance on LIBERO and RoboTwin 2.0, demonstrates exceptional out-of-distribution generalization on LIBERO-Plus, and achieves end-to-end inference in just 66.5 ms, yielding a 3.2× speedup over Fast-WAM.

Abstract

World Action Models (WAMs) couple robot action prediction with video world models. Existing WAMs with shared-backbone and Mixture-of-Transformers designs generally tie the depth of the action module to that of the video backbone, resulting in substantial computational overhead and high inference latency. To address this limitation, we introduce Dock of Transformer (DoT), a video-centric design principle that treats a pretrained video Transformer as a representation hub and connects lightweight output-heads through docking interfaces. This enables flexible output-head design while providing direct access to representations from all layers of the backbone. We then introduce \textbf{Faster-WAM}, an instantiation of DoT for WAMs, which docks a single-layer action head onto a 30-layer video backbone. The docking interface fuses keys and values from all video layers and applies RoPE realignment. Without additional embodied pretraining, Faster-WAM achieves competitive performance on LIBERO and RoboTwin 2.0 while demonstrating strong out-of-distribution generalization on LIBERO-Plus. Faster-WAM also achieves the lowest end-to-end latency in our controlled comparison, requiring only 66.5 ms per inference --- a \(3.2\times\) speedup over Fast-WAM. Overall, these results demonstrate that the video-centric DoT architecture supports flexible task-specific head design while delivering low inference latency, strong action-prediction performance, and robust generalization.

Problem

• World Action Models
• action prediction
• video backbone
• inference latency
• computational overhead

Innovation

• Dock of Transformer
• World Action Models
• lightweight action head
• video-centric architecture
• low-latency inference