aiqbench analysis
DynamicWAM advances robotics by introducing dual-path motion conditioning into World-Action Models, enabling robots to track moving targets with superior temporal awareness and responsiveness compared to synchronous baselines. The architecture leverages frozen video VAEs for optical flow history while injecting kinematic descriptors directly into the action expert, achieving significant performance gains on both simulated DOMINO benchmarks and diverse real-world manipulation tasks. This work represents a notable step forward in making large-scale world models viable for dynamic environments where target motion is critical to success.
DynamicWAM: Dual-Path Motion Conditioning for World-Action Models in Dynamic Manipulation
arXiv:2608.00793v1 Announce Type: new Abstract: Dynamic manipulation requires robots to infer target motion and respond promptly, yet existing World-Action Models (WAMs) typically condition only on the current frame and execute large backbones synchronously, limiting motion awareness and responsive control in dynamic scenes. We propose DynamicWAM, a compact WAM for dynamic object manipulation with dual-path motion conditioning. DynamicWAM introduces history-flow conditioning, encoding temporally aligned optical-flow frames alongside the current observation through a frozen pretrained video VAE to preserve spatial motion structure, while injecting kinematic descriptors of displacement, duration, velocity, and acceleration into the action expert to provide motion magnitude and timing. The two complementary paths are fused through joint world-action attention. A distilled compact backbone and real-time chunking (RTC)-based asynchronous execution further enable responsive control. On DOMINO, DynamicWAM achieves a 38.2% success rate and a 53.2 manipulation score, outperforming all evaluated baselines. Across 12 real-world tasks spanning linear, circular, and compound target motion, it achieves a 46.7% average success rate, exceeding the strongest baseline by 22.9 percentage points.
Abstract:Dynamic manipulation requires robots to infer target motion and respond promptly, yet existing World-Action Models (WAMs) typically condition only on the current frame and execute large backbones synchronously, limiting motion awareness and responsive control in dynamic scenes. We propose DynamicWAM, a compact WAM for dynamic object manipulation with dual-path motion conditioning. DynamicWAM introduces history-flow conditioning, encoding temporally aligned optical-flow frames alongside the current observation through a frozen pretrained video VAE to preserve spatial motion structure, while injecting kinematic descriptors of displacement, duration, velocity, and acceleration into the action expert to provide motion magnitude and timing. The two complementary paths are fused through joint world-action attention. A distilled compact backbone and real-time chunking (RTC)-based asynchronous execution further enable responsive control. On DOMINO, DynamicWAM achieves a 38.2% success rate and a 53.2 manipulation score, outperforming all evaluated baselines. Across 12 real-world tasks spanning linear, circular, and compound target motion, it achieves a 46.7% average success rate, exceeding the strongest baseline by 22.9 percentage points.
18 pages, 9 figures. Project page: this https URL
Robotics (cs.RO)
arXiv:2608.00793 [cs.RO]
(or arXiv:2608.00793v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2608.00793 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 264 words
Related coverage
NVIDIA Releases Cosmos 3: A Two-Tower Mixture-of-Transformers Foundation Model Unifying Physical Reasoning, World Generation, and Action Generation
MarkTechPost · Jun 3, 2026
Game Design Creativity evals
LessWrong · Oct 12, 2026
ForgeWM: Progressive Causal Training for Few-Step Action-Conditioned Video World Models
arXiv · Aug 17, 2026
How to Evaluate General-Purpose Robot Policies for Real-World Deployment
NVIDIA · Jul 12, 2026