Valdi: Value Diffusion World Models

arXiv 2026 42.8 method

TLDR

Valdi combines latent diffusion dynamics with MPC for online planning, matching deterministic baselines on CarRacing while exposing multimodality trade-offs.

Reasoning

Strengths include a novel integration of diffusion models with MPC for low-latency planning and end-to-end training. Weaknesses are preliminary experiments on a simple environment (CarRacing) and limited evaluation of the multimodality trade-off, with no real-world or complex benchmarks.

Read-first score

Read-first score 42.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 35.

Recency 6%

100

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

Methodology quality 18%

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

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

Reproducibility 18%

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

Citation impact 18%

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

Citation velocity 12%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 486.

Keyword Scores

world model

world dynamics prediction

model-based reinforcement learning world model

generative world model

world simulator

interactive world model

video world model

Deep Analysis

Innovations

Combines end-to-end online training for Model Predictive Control with a latent diffusion dynamics model (Valdi).
Uses a single diffusion step at both training and inference to enable low-latency latent planning with diffusion models.
Identifies and exposes a trade-off between predictive multimodality and control performance in this setup.

Methodology

Valdi integrates a latent diffusion dynamics model into Model Predictive Control and is trained end-to-end online. The method is evaluated on the CarRacing environment, using a single diffusion step at training and inference, and compared against a deterministic MLP baseline.

Key Results

Valdi with a single diffusion step matches the performance of a deterministic MLP baseline on CarRacing, and experiments reveal a trade-off between predictive multimodality and control performance.

Limitations

Evaluation is limited to preliminary experiments on the CarRacing environment only.
A trade-off exists between predictive multimodality and control performance, which may limit benefits of diffusion-based multimodality.
Using a single diffusion step may restrict the model's ability to capture complex multimodal futures compared to full iterative inference.

Resources

Paper