Discrete-WAM: Unified Discrete Vision-Action Token Editing for World-Policy Learning

arXiv 2026 63.1 method, application

TLDR

Discrete-WAM unifies vision and action tokens for world-policy learning in autonomous driving, enabling joint world modeling and policy generation via hierarchical token editing.

Reasoning

The paper introduces a novel discrete token alignment approach that jointly trains world and policy models, showing strong empirical results on autonomous driving benchmarks. However, the evaluation is limited to a single domain and lacks comparison with model-based RL methods, and the abstract does not discuss potential limitations or failure cases.

Read-first score

Read-first score 63.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 44.

Recency 6%

100

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

Citation impact 18%

91.3

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.91320217

Methodology quality 18%

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

Topical relevance 29%

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

Reproducibility 18%

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

Citation velocity 12%

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

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 386.

Keyword Scores

world model

world dynamics prediction

generative world model

interactive world model

video world model

model-based reinforcement learning world model

world simulator

Deep Analysis

Innovations

Unified discrete vision-action token space representing visual observations, future states, high-level decisions, and ego actions
Multi-task and multi-stage pretraining jointly training world modeling, world-policy modeling, and policy modeling
Hierarchical decision prediction and parallel action-token editing with confidence-based scheduling for downstream planning

Methodology

Discrete-WAM encodes visual observations, future states, high-level decisions, and ego actions into a shared discrete token space. It then performs multi-task and multi-stage pretraining to jointly learn world modeling, world-policy modeling, and policy modeling, enabling action-conditioned future prediction to support policy generation. For planning, it decomposes policy generation into hierarchical decision prediction and parallel action-token editing, where a decision token provides a high-level skeleton and confidence-based scheduling refines dense future actions.

Key Results

On large-scale autonomous-driving benchmarks, Discrete-WAM achieves strong planning performance while supporting controllable future generation, counterfactual evaluation, surprise-based world-model analysis, and efficient parallel policy decoding.

Resources

Paper