Fast-WAM: Do World Action Models Need Test-time Future Imagination?

TL;DR

Fast-WAM demonstrates that explicit future imagination at test time is not necessary for strong embodied control, as training with video modeling alone suffices, leading to faster and competitive performance.

Contribution

The paper introduces Fast-WAM, a WAM architecture that removes test-time future prediction, showing that training with video modeling alone maintains performance and significantly improves speed.

Findings

Fast-WAM achieves over 4x faster inference than traditional WAMs.

Video co-training during training is more critical than test-time future prediction.

Fast-WAM performs competitively on simulation and real-world benchmarks.

Abstract

World Action Models (WAMs) have emerged as a promising alternative to Vision-Language-Action (VLA) models for embodied control because they explicitly model how visual observations may evolve under action. Most existing WAMs follow an imagine-then-execute paradigm, incurring substantial test-time latency from iterative video denoising, yet it remains unclear whether explicit future imagination is actually necessary for strong action performance. In this paper, we ask whether WAMs need explicit future imagination at test time, or whether their benefit comes primarily from video modeling during training. We disentangle the role of video modeling during training from explicit future generation during inference by proposing \textbf{Fast-WAM}, a WAM architecture that retains video co-training during training but skips future prediction at test time. We further instantiate several Fast-WAM variants to enable a controlled comparison of these two factors. Across these variants, we find that Fast-WAM remains competitive with imagine-then-execute variants, while removing video co-training causes a much larger performance drop. Empirically, Fast-WAM achieves competitive results with state-of-the-art methods both on simulation benchmarks (LIBERO and RoboTwin) and real-world tasks, without embodied pretraining. It runs in real time with 190ms latency, over 4$\times$ faster than existing imagine-then-execute WAMs. These results suggest that the main value of video prediction in WAMs may lie in improving world representations during training rather than generating future observations at test time. Project page: https://yuantianyuan01.github.io/FastWAM/