HiLo-WAM: Decoupling Long-Horizon Task Evolution and Local Dynamics in World Action Models
HiLo-WAM: Decoupling Long-Horizon Task Evolution and Local Dynamics in World Action Models
This research paper introduces HiLo-WAM, a novel World Action Model (WAM) architecture designed to improve robotic manipulation by decoupling long-horizon task evolution from local visual dynamics. The model employs a dual-stream approach with a sparse long-horizon stream for historical context and a dense local stream for short-horizon interactions, integrated within a unified dual-DiT framework. Experimental results on RoboTwin 2.0 show an average success rate of 93.70%, while evaluations on RoboDojo yield a progress score of 17.43. The method also demonstrates robust performance on the Fourier GR-3 robot and achieves a 2.93x speedup through progressive inference techniques.
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
progress score
17.43
View original evidence
e on RoboTwin 2.0 and a 17.43 progress score on RoboDojo long-horizon tasks, with robust in-domain and OOD performance on the Fourier GR-3 robot. Ablations confirm the complementary contributions of long-horizon task evolution and local interaction dynamics, while temporal-scale analyses highlight the importance of balancing temporal coverage and resolution. Progressive inference further achieves a 2.93x speedup whil
Open source S5speedup factor
2.93
View original evidence
e on RoboTwin 2.0 and a 17.43 progress score on RoboDojo long-horizon tasks, with robust in-domain and OOD performance on the Fourier GR-3 robot. Ablations confirm the complementary contributions of long-horizon task evolution and local interaction dynamics, while temporal-scale analyses highlight the importance of balancing temporal coverage and resolution. Progressive inference further achieves a 2.93x speedup whil
Open source S5
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
eractions. The two streams operate at different temporal scales within a unified dual-DiT architecture and guide actions through structured cross-stream attention. Progressive inference further reduces unnecessary computation through stream-wise partial denoising, joint denoising, and dynamic skipping. Experiments demonstrate strong long-horizon manipulation performance: HiLo-WAM achieves a 93.70% average success rat
Open source S4
e on RoboTwin 2.0 and a 17.43 progress score on RoboDojo long-horizon tasks, with robust in-domain and OOD performance on the Fourier GR-3 robot. Ablations confirm the complementary contributions of long-horizon task evolution and local interaction dynamics, while temporal-scale analyses highlight the importance of balancing temporal coverage and resolution. Progressive inference further achieves a 2.93x speedup whil
Open source S5
HiLo-WAM: Decoupling Long-Horizon Task Evolution and Local Dynamics in World Action Models
Open source S1
arXiv:2611.05023v1 Announce Type: new Abstract: World Action Models (WAMs) improve robotic manipulation by jointly modeling visual dynamics and robot actions. However, most existing WAMs model dense short-horizon visual dynamics, capturing local interactions but providing limited long-range task context. Extending prediction horizons increases uncertainty and computation while entangling task-level evolution with fin
Open source S2
e-grained local dynamics, obscuring their dependencies and distinct roles in action generation. We propose HiLo-WAM, which decouples long-horizon task evolution from local visual dynamics while jointly modeling their interaction with robot actions. A sparse long-horizon visual stream captures historical task context and future evolution, while a dense local dynamics visual stream models short-horizon robot-object int
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org