UMI-WM: Interactive World Modeling for Biman-ual Manipulation from Wrist Cameras
UMI-WM: Interactive World Modeling for Bimanual Manipulation from Wrist Cameras
UMI-WM is an action-conditioned bimanual world model that jointly generates both wrist views from observation history, per-arm relative camera poses, and gripper widths, with cross-view consistency, action following, and visual fidelity. The model outperforms baselines in action following and generates paired wrist videos 6.9 times faster than the Teacher model.
The paper introduces UMI-WM, a bimanual world model that generates consistent wrist-view videos from observation history and per-arm camera poses, addressing challenges in maintaining cross-view consistency and action fidelity in manipulation tasks.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
generated videos
6.9
View original evidence
following, while the four-step Student generates paired wrist videos 6.9 times faster than the Teacher.
Open source S6denoising steps
4
View original evidence
ware geometry conditioning and camera-guided noise further support per-view motion control and cross-view geometric exchange. We adapt the model for block-causal generation and distill the 30-step Teacher into a Student with four denoising steps. We train UMI-WM on self-collected open-world and in-studio UMI recordings combined with public manipulation data. To evaluate action following, we use an inverse dynamics mo
Open source S4teacher steps
30
View original evidence
ware geometry conditioning and camera-guided noise further support per-view motion control and cross-view geometric exchange. We adapt the model for block-causal generation and distill the 30-step Teacher into a Student with four denoising steps. We train UMI-WM on self-collected open-world and in-studio UMI recordings combined with public manipulation data. To evaluate action following, we use an inverse dynamics mo
Open source S4reference-video fidelity metrics
5
View original evidence
del (IDM) to recover relative camera poses and gripper widths from generated videos, and introduce Bimanual Action Fidelity (BAF) to measure how closely these recovered actions match the conditioning actions. UMI-WM achieves the best results on all five reference-video fidelity metrics and the best or competitive performance on VBench and WorldArena. The Teacher and Student outperform all compared baselines in action
Open source S5
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
following, while the four-step Student generates paired wrist videos 6.9 times faster than the Teacher.
Open source S6
ware geometry conditioning and camera-guided noise further support per-view motion control and cross-view geometric exchange. We adapt the model for block-causal generation and distill the 30-step Teacher into a Student with four denoising steps. We train UMI-WM on self-collected open-world and in-studio UMI recordings combined with public manipulation data. To evaluate action following, we use an inverse dynamics mo
Open source S4
del (IDM) to recover relative camera poses and gripper widths from generated videos, and introduce Bimanual Action Fidelity (BAF) to measure how closely these recovered actions match the conditioning actions. UMI-WM achieves the best results on all five reference-video fidelity metrics and the best or competitive performance on VBench and WorldArena. The Teacher and Student outperform all compared baselines in action
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org