AR Forcing: Towards Long-Horizon Robot Navigation World Model

arXiv 2026 62.1 method

TLDR

AR Forcing trains diffusion navigation world models autoregressively to reduce distribution shift, improving long-horizon prediction consistency and trajectory accuracy.

Reasoning

Strengths: Addresses a key distribution shift problem in diffusion-based navigation world models with a simple, plug-in training strategy that requires no extra losses or discriminators, validated on multiple real-world datasets. Weaknesses: Limited to robot navigation domain; no comparison to non-diffusion methods or analysis of computational overhead; code not yet released.

Read-first score

Read-first score 62.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 43.

Recency 6%

100

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

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.89965508

Methodology quality 18%

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

Topical relevance 29%

61.4

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,dataset

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: 380.

Keyword Scores

world model

generative world model

world dynamics prediction

video world model

interactive world model

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

AR Forcing: an autoregressive training strategy for diffusion-based robot navigation world models that integrates standard diffusion loss into the autoregressive training loop.
The method uses the model's own predictions to update context and optimize the single step noise prediction objective, explicitly exposing the model to inference state distribution during training.
No additional discriminators or distribution-matching losses are required; the original diffusion framework and sampler are retained.

Methodology

AR Forcing trains a diffusion model autoregressively: at each step, the model uses its own predictions to update the context and optimizes the single step noise prediction objective using the standard diffusion loss. This exposes the model to the inference state distribution during training, addressing the distribution shift between parallel supervision training and autoregressive inference. The method retains the original diffusion framework and sampler without requiring additional discriminators or distribution-matching losses.

Key Results

On multi-domain navigation datasets (RECON, SCAND, HuRoN, TartanDrive), AR Forcing improved the consistency of generated images during long-horizon navigation and the accuracy of predicted trajectories compared to strong baselines, enhancing robustness in complex known and unknown environments.

Resources

Paper