DeltaWAM: Delta World Action Models for Bimanual Manipulation
- 类型:arxiv
- 标识:2609.28811
- 链接:https://arxiv.org/abs/2609.28811
- 主分类:multimodal
- 形态:method
- TLDR:World-action models (WAMs) transfer visual and motion priors from pretrained video generators to robot control by jointly modeling visual dynamics and actions. Existing WAMs, however, predict dense future frames during training, repeatedly modeling largely unchanged content and coupling action-conditioned dynamics to nuisance appearance variations. At inference, processing each complete observation with the heavy video expert bottlenecks few-step action generation. Accordingly, we propose DeltaWAM, which jointly predicts visual deltas and actions using dense-anchor, sparse-delta, and action st
- 副分类:engineering
- 待LLM分类:否
- 来源文件:
- /inbox/tom/_candidates/2026-09-25-agent-rag-longcontext-candidates.json