Newest papers

De-Diffusing WAMs: Action-Sufficient Direct Prediction for Robot Control

Shaoyi Zheng, Zhongtao Jiang, Yue Peng, Wenbo Lu, Shunchi Zhang, Xi Wang, Shengji Wan · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.28122 · cited by 0 works (OpenAlex) · open-access PDF

Portuguese

Esses modelos de ação de mundo latente (WAMs) usam futuros visuais previsíveis para condicionar ações de robôs, mas herdam inferências iterativas dispendiosas de especialistas treinados em difusão. Perguntamos o que um futuro previsível de um único passo deve preservar para apoiar previsões de ação de um único passo precisas. Na WAM avaliada, a reamostragem de ruído de vídeo tem pouco efeito nas ações de baixo nível, mas o truncating ingênuo de ambos os fluxos para um passo reduz substancialmente o sucesso da tarefa. A supervisão de espaço de vídeo preenche apenas parcialmente essa lacuna, e uma análise de sensibilidade local mostra que os erros de previsão de visão têm consequências altamente desiguais para a precisão da ação. Esses resultados motivam a DirectWAM, que

Machine translation (Meta M2M100); the English original follows.

Abstract

Cascade world-action models (WAMs) use predicted visual futures to condition robot actions, but inherit costly iterative inference from diffusion-pretrained experts. We ask what a single-step predicted future must preserve to support accurate single-step action prediction. In the evaluated WAM, resampling video noise has little effect on downstream actions, yet naively truncating both streams to one step substantially reduces task success. Video-space supervision only partially closes this gap, and a local sensitivity analysis shows that vision prediction errors have highly unequal consequences for action accuracy. These findings motivate DirectWAM, which converts pretrained video and action experts into single-step predictors without stochastic sampling through action-sufficient distillation, while retaining the explicit future-to-action cascade. After learning a one-step action model, we train the future predictor by decoding actions from its own predicted futures and backpropagating demonstration action error through the expert, with an optional future-latent reconstruction loss. The resulting objective evaluates the student-generated future through its downstream control consequences rather than relying on representation matching alone. DirectWAM achieves 97.1% average success across four LIBERO suites and 85.4% on RoboTwin 2.0, outperforming Flash-WAM at the same one-evaluation-per-stream budget. This advantage has also extended to three real-world tasks that we designed. Compared with the full iterative teacher, it reduces inference latency by $28.0\times$.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Artificial intelligence, Inference, Machine learning, Task (project management), Action (physics)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange