dWorldEval: Scalable Robotic Policy Evaluation via Discrete Diffusion World Model

arXiv 2026 52.1 method, application

TLDR

Proposes dWorldEval, a discrete diffusion world model for scalable robotic policy evaluation using unified token space and progress token.

Reasoning

The paper introduces a novel discrete diffusion world model that unifies vision, language, and actions into tokens, with a progress token for task completion. Strengths include strong empirical results on multiple benchmarks and real-robot tasks. Weaknesses are not explicitly discussed in the abstract, but the approach appears promising for scalable evaluation.

Read-first score

Read-first score 52.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Recency 6%

100

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

Citation impact 18%

64.7

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.64694858

Topical relevance 29%

64.3

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: 308.

Keyword Scores

world model

world dynamics prediction

world simulator

generative world model

video world model

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Mapping all modalities (vision, language, robotic actions) into a unified token space
Single transformer-based denoising network for discrete diffusion world model
Sparse keyframe memory to maintain spatiotemporal consistency
Progress token indicating task completion degree for automatic success detection

Methodology

dWorldEval uses a discrete diffusion world model as a scalable evaluation proxy. It maps vision, language, and robotic actions into a unified token space and denoises them with a single transformer network. A sparse keyframe memory ensures spatiotemporal consistency, and a progress token is introduced to automatically determine task success when its value reaches 1.

Key Results

dWorldEval significantly outperforms previous approaches (WorldEval, Ctrl-World, WorldGym) on LIBERO, RoboTwin, and multiple real-robot tasks.

Resources

Paper