Yude Zou, Yifei Yao, Guanbo Jiao, Jianming Ma, Guanjie Zheng, Xing Gao · 2026 · arXiv
DOI: 10.48550/arXiv.2610.26813 · open-access PDF
Os modelos de ação mundial (WAMs) preveem conjuntamente o vídeo futuro e geram ações, mas permanecem sensíveis a perturbações visuais irrelevantes à tarefa, ou seja, mudanças na aparência visual que não devem afetar o comportamento do robô. Identificamos uma lacuna no treinamento convencional WAM: a previsão de vídeo incentiva as representações a preservarem informações de aparência, enquanto a geração de ação não impõe explicitamente a invariância à aparência irrelevante à tarefa.
Machine translation (Meta M2M100); the English original follows.
World Action Models (WAMs) jointly predict future video and generate actions, yet remain sensitive to task-irrelevant visual perturbations, i.e., changes in visual appearance that should not affect robot behavior. We identify a gap in conventional WAM training: video prediction encourages representations to preserve appearance information, while action generation does not explicitly enforce invariance to task-irrelevant appearance. Data augmentation exposes WAMs to such perturbations but does not require perturbed and nominal observations to produce consistent action representations, and align
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.26813, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange