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

Spanish

Estos modelos de acción del mundo en cascada (WAMs) utilizan futuros visuales predecibles para condicionar las acciones de los robots, pero heredan una inferencia iterativa costosa de expertos pretrained por difusión. Preguntamos qué debe conservar un futuro predecible de un solo paso para apoyar una predicción exacta de acción de un solo paso. En WAM evaluado, el ruido de reemplazo tiene poco efecto en las acciones a continuación, pero el truncating ingenuo de ambos flujos a un paso reduce sustancialmente el éxito de la tarea. La supervisión del espacio de vídeo sólo cierra parcialmente esta brecha, y un análisis de sensibilidad local muestra que los errores de predicción de la visión tienen consecuencias muy desiguais para la exactitud de la acción. Estos hallazgos motivan DirectWAM

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