OmniNWM: Omniscient Driving Navigation World Models

arXiv 25.10 2025 52.8 method, application

TLDR

OmniNWM is a panoramic navigation world model that generates multi-modal videos, enables precise action control, and provides dense rewards for closed-loop evaluation of driving agents.

Reasoning

Strengths include a unified probabilistic framework addressing state, action, and reward dimensions, zero-shot trajectory control across datasets, and long-horizon generation via structured panoramic forcing. Weaknesses are domain specificity to driving, reliance on an in-house dataset limiting reproducibility, and no discussion of real-time performance.

Read-first score

Read-first score 52.8, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 59.

Recency 6%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Topical relevance 29%

84.3

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=dataset,evaluation,experiment,metric

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,dataset,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: 674.

Keyword Scores

world model

generative world model

video world model

world simulator

interactive world model

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Panoramic video generation with pixel-aligned RGB, semantics, metric depth, and 3D occupancy via joint distribution modeling
Structured panoramic forcing strategy to mitigate autoregressive exposure bias and stabilize long-horizon generation via stochastic manifold thickening
Canonical geometric action encoding using normalized panoramic Plücker ray-maps for zero-shot trajectory control across heterogeneous datasets and camera configurations
Intrinsic occupancy-grounded dense rewards derived from generated 3D volumes for closed-loop simulation and planning agent evaluation

Methodology

OmniNWM is a consistent probabilistic framework that addresses state, action, and reward dimensions for autonomous driving world models. It generates panoramic multi-modal videos (RGB, semantics, depth, 3D occupancy) with pixel-level alignment, uses a structured panoramic forcing strategy to reduce autoregressive drift, encodes actions via canonical Plücker ray-maps for decoupled motion dynamics, and derives dense rewards from generated 3D occupancy volumes.

Key Results

OmniNWM achieves state-of-the-art performance in generation fidelity and control precision, demonstrating remarkable zero-shot robustness to novel scenes on NuPlan and in-house datasets with distinct camera rigs.

Resources

Paper Project