AHA-WAM:Asynchronous Horizon-Adaptive World-Action Modeling with Observation-Guided Context Routing
Summary
AHA-WAM is an asynchronous world-action model that uses dual Diffusion Transformers to decouple world prediction from action execution, achieving efficient long-horizon planning and real-time control. It achieves state-of-the-art performance on robotic manipulation tasks with up to 92.8% success on RoboTwin and 78.3% on real-world tasks, while reaching 24.17 Hz closed-loop control.
View Cached Full Text
Cached at: 06/09/26, 08:41 AM
Paper page - AHA-WAM:Asynchronous Horizon-Adaptive World-Action Modeling with Observation-Guided Context Routing
Source: https://huggingface.co/papers/2606.09811 Authors:
,
,
,
,
,
,
,
,
,
,
,
Abstract
AHA-WAM is an asynchronous world-action model that uses dual Diffusion Transformers to enable efficient long-horizon planning and real-time action execution in robotic manipulation tasks.
World-action modelshave emerged as a promising paradigm for robot manipulation, jointly modeling visual scene dynamics and actions to inject physical priors into policy learning. However, existingworld-action modelscouple world prediction and action execution at the same temporal resolution, forcing the world branch to model near-term frame variations that are redundant and weakly informative. We posit that strictly binding world prediction and action execution to the same temporal rhythm may underutilize the potential of the video branch for embodied control. Therefore, we propose AHA-WAM, an Asynchronous Horizon-Adaptive World-Action Model built on adual Diffusion Transformer(DiT) architecture that reorganizes world-action modeling around this temporal asymmetry. AHA-WAM instantiates thevideo DiTas a low-frequencyworld plannerthat maintains rolling key-value memory over past observations and exposes reusable layerwise latent context encoding long-horizon scene evolution, while a high-frequencyaction DiTexecutes short action chunks in closed loop by querying this context throughlayerwise joint attention. To support asynchronous execution, we introducehorizon-adaptive offset trainingandObservation-Guided Video-Context Routing(OVCR), which together let the action expert exploit long-horizon world context while remaining responsive to real-time execution state without rerunning thevideo DiT. Experiments on RoboTwin and real-world manipulation tasks show that AHA-WAM achieves state-of-the-art performance without any robot-data pretraining, attaining 92.80% average success on RoboTwin and 78.3% success across 4 real-world tasks, while reaching 24.17 Hzclosed-loop controlwith a 4.59x speedup over Fast-WAM.
View arXiv pageView PDFProject pageAdd to collection
Get this paper in your agent:
hf papers read 2606\.09811
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2606.09811 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2606.09811 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2606.09811 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
RealtimeWAM: One-Step Asynchronous World Action Models
RealtimeWAM proposes one-step asynchronous World Action Models using Teacher-Anchored Consistency Distillation (TACD) and Cross-Expert Wavefront Pipelining (CEWP), achieving near-lossless performance (<1% drop) on LIBERO, LIBERO-Plus, and RoboTwin with up to 25x end-to-end speedup on H100.
Light-WAM: Efficient World Action Models with State-Fusion Action Decoding
Light-WAM is a lightweight world action model for efficient robot manipulation that uses a compact video backbone and downsampled latent space for future-video supervision, achieving high performance with low inference latency.
Agentic World Action Models
The paper introduces AW-0, an agentic world action model framework that unifies code-based planning with joint video and action prediction in a mixture-of-transformers architecture, enabling reusable skill libraries for robot manipulation. Evaluations on RoboTwin2.0 and bimanual manipulation tasks show improved performance over VLA and WAM baselines.
World Action Modeling with Progressive Visual Planning
ProWAM is a progressive world action model that jointly predicts robot actions and an ordered sequence of sparse visual sub-goals, enabling efficient long-horizon control without dense video generation. It achieves state-of-the-art results on LIBERO-Plus and RoboTwin, plus 70% zero-shot real-world success.
Rolling-WAM: World Action Models with Rolling Imagination
Rolling-WAM introduces a method to distribute denoising across replanning cycles in world action models for robotic manipulation, achieving a 4.5x speedup in replanning while maintaining competitive performance.