Efficient-WAM: A 1B-Parameter World-Action Model with Low-Cost Future Imagination

2026-06-08

AI Summary

This work addresses the high computational cost and inference latency of existing world-action models, which rely on high-fidelity future video prediction and thus hinder real-time deployment. To overcome this limitation, the authors propose a lightweight world model that treats future video prediction as a guiding signal rather than a reconstruction target. The approach leverages a compact video expert adapted from WAN-2.2-5B, employs token-sparse latent video representations, and introduces an asymmetric video-action denoising mechanism to substantially reduce computational overhead. Evaluated on RoboTwin 2.0 and real-world manipulation tasks, the method maintains strong control performance while achieving an inference latency of approximately 100 milliseconds on a single GPU—representing a 30-fold speedup over current world-action models (WAMs).

Abstract

World-Action Models (WAMs) have emerged as a promising paradigm for embodied control by coupling future visual prediction with action generation. However, most existing WAMs rely on photorealistic future prediction, which incurs high inference latency and makes real-time robot deployment difficult. This motivates a more efficient WAM design that preserves the control benefits of future visual prediction while reducing its inference cost. We introduce Efficient-WAM, a World-Action Model that reduces the cost of future imagination while preserving its control benefit. Efficient-WAM improves inference efficiency via a compact video expert transferred from WAN-2.2-5B, token-sparse video latents, and asymmetric video-action denoising that allocates fewer sampling steps to video than to actions. Instead of optimizing the future branch for visual fidelity, Efficient-WAM treats future video prediction as a compact guidance signal for action generation. Comprehensive experiments on RoboTwin 2.0 and real-world manipulation tasks show that Efficient-WAM maintains strong action performance despite visibly coarse future predictions. While maintaining competitive control capabilities, our 1B-parameter model can reduce per-chunk latency to around 100 ms during physical deployment, achieving a 30x speedup over existing WAMs.

Problem

• World-Action Models
• future visual prediction
• inference latency
• real-time robot deployment
• embodied control

Innovation

• World-Action Model
• efficient inference
• future imagination
• asymmetric denoising
• video-action generation