Agentic World Action Models

arXiv 2026 36.3 method

TLDR

Introduces AW-0, an agentic world action model combining code planning, future video, and robot action prediction via a mixture-of-transformers, improving manipulation tasks and skill reuse.

Reasoning

The paper presents a novel integration of code, video, and action generation in a unified architecture, with a curation pipeline for reusable skills and evaluation on simulation and physical bimanual tasks. However, the abstract omits quantitative results and specific limitations, making it hard to fully assess the claimed improvements.

Read-first score

Read-first score 36.3, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 44.

Recency 5%

100

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

Topical relevance 25%

62.9

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 15%

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

Reproducibility 15%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 10%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 355.

Keyword Scores

world model

generative world model

video world model

world dynamics prediction

interactive world model

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

Introduces AW-0, an agentic world action model framework that integrates code-based planning with joint video and action prediction.
Proposes a unified mixture-of-transformers architecture that jointly models code, future video, and robot actions as three coupled generative streams.
Uses code to serve two complementary roles: structuring multi-step prediction and control within the world model, and representing skills as reusable programs for agentic execution.
Develops an automated curation pipeline that labels robot and human video trajectories with aligned programs and consolidates recurring routines into reusable skill libraries.

Methodology

AW-0 jointly models code, future video, and robot actions as three coupled generative streams within a unified mixture-of-transformers architecture. A code expert generates parameterized skill routines and control logic, while video and action experts generate future visual states and low-level motor commands. An automated curation pipeline labels robot and human video trajectories with aligned programs and builds reusable skill libraries from recurring routines.

Key Results

Evaluations on RoboTwin 2.0 and physical bimanual manipulation tasks show improved task performance over VLA and WAM baselines, with additional gains in language following, semantic reasoning, and robustness.

Resources

Paper