English(EN)DUET-DINO: Simultaneous Cross-View World Modeling for Latent Planning in Robot Manipulation
新的AI模型通过改进的记忆和规划能力推动机器人操作发展
作者PulseAugur 编辑部·[5 个来源]·
研究人员开发了新的机器人操作方法，提高了在长时任务中的性能。2AM系统将任务记忆与动作模型分离，实现了更精确的控制，并在LIBERO-Mem数据集上比现有基线提高了61.5分。DUET-DINO是另一项进展，它使用同时跨视图世界建模来增强潜在规划，实现完整的7-DoF末端执行器控制，在到达任务上以高达92%的成功率超越了单视图模型。GE-Act 2.0专注于从头开始预训练和扩展世界-动作模型，通过更多的训练数据展示了显著的成功率提升，并显示出强大的零样本泛化能力。
AI
arXiv:2609.11308v1 Announce Type: cross Abstract: Long-horizon robot manipulation requires memory, but not necessarily inside the action policy. To address such tasks, current agentic systems often combine VLAs with planners and geometric tools, sometimes using additional depth o…
Long-horizon robot manipulation requires memory, but not necessarily inside the action policy. To address such tasks, current agentic systems often combine VLAs with planners and geometric tools, sometimes using additional depth or calibrated geometry. These systems confound attr…
Action-conditioned latent world models predict future visual representations, enabling zero-shot goal-conditioned robot planning and control. However, their predictions for fine-grained spatial and rotational actions are unreliable for full 7-DoF end-effector control. To address …
GE-Act 2.0 is a world-action model trained from scratch with a control-oriented autoencoder, single-step visual planner, and inverse dynamics model, using knowledge-aligned selective optimization to enable scalable zero-shot robot manipulation across diverse skills and conditions…
arXiv cs.CV
TIER_1English(EN)·Nisarga Nilavadi, Ralf R\"omer, Moritz Reuss, Michael Krawez, Tobias J\"ulg, Angela P. Schoellig, Rudolf Lioutikov, Wolfram Burgard·
arXiv:2609.10506v1 Announce Type: cross Abstract: Action-conditioned latent world models predict future visual representations, enabling zero-shot goal-conditioned robot planning and control. However, their predictions for fine-grained spatial and rotational actions are unreliabl…
