HarmoWAM: Harmonizing Generalizable and Precise Manipulation via Adaptive World Action Models

arXiv 2026 52.2 method

TLDR

HarmoWAM unifies predictive and reactive control via a world model with adaptive gating for generalizable and precise robot manipulation.

Reasoning

The paper clearly identifies a trade-off between two WAM paradigms and proposes a novel architecture with predictive and reactive experts and a gating mechanism. However, the abstract lacks explicit mention of real-world experiments or benchmarks, and the evaluation details are not provided, limiting assessment of empirical validity.

Read-first score

Read-first score 52.2, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 42.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

72.6

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.72616031

Topical relevance 29%

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=evaluation,experiment

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FrontierBridge

Rank sensitivity

Stability: volatile; rank range: 279.

Keyword Scores

world model

world dynamics prediction

video world model

interactive world model

model-based reinforcement learning world model

generative world model

world simulator

Deep Analysis

Innovations

Observation of fundamental trade-off between Imagine-then-Execute and Joint Modeling paradigms in World Action Models
Proposal of HarmoWAM, an end-to-end WAM that unifies predictive and reactive control via a world model
Design of two complementary action experts: predictive expert using latent dynamics for iterative action generation, and reactive expert inferring actions from predicted visual evolution
Process-Adaptive Gating Mechanism for adaptive coordination between experts
Demonstration of strong zero-shot generalization across six real-world robotic tasks with variations in background, position, and object semantics, outperforming prior VLA and WAM models by 33% and 29%

Methodology

HarmoWAM uses a world model to provide spatio-temporal physical priors that condition two complementary action experts: a predictive expert leveraging latent dynamics for iterative action generation, and a reactive expert directly inferring actions from predicted visual evolution. A Process-Adaptive Gating Mechanism automatically determines the timing and location of switching between these experts. The model is evaluated on six real-world robotic tasks across three training-unseen test environments covering variations in background, position, and object semantics.

Key Results

HarmoWAM achieves strong zero-shot generalization across unseen test environments, significantly outperforming prior state-of-the-art VLA models and WAMs by margins of 33% and 29%, respectively.

Resources

Paper