SPW-Nav: A Streaming Panoramic World Model for Language-Guided Navigation

arXiv 2026 37.6 method

TLDR

SPW-Nav is a streaming panoramic world model that generates real-time 2K 360-degree video from language navigation instructions and a new dataset.

评分理由

The paper introduces a novel approach with spherical rotation decoupling, pose-aligned conditioning, and multi-term memory for long-span panoramic generation, and includes a new dataset. Weaknesses include limited information about quantitative evaluation details and broader benchmark comparisons in the abstract.

Read-first 评分解释

综合优先阅读分 37.6，由主题、引用、图谱、方法、可复现性和近期性等信号加权得到。原始总分保留为 52。

近期性 5%

100

使用温和的时间衰减，让近期论文更容易浮现，同时保留较早基础工作的价值。年份：2026

主题相关性 25%

74.3

使用现有 LLM 关键词相关性评分，并归一化到 0-100。关键词：world model、world simulator、generative world model、interactive world model、video world model、world dynamics prediction、model-based reinforcement learning world model

方法质量 15%

检查可见的摘要与分析字段，寻找实验、数据集、基线、指标和局限性等方法证据。命中信号：数据集

可复现性 15%

检查链接和可见文本中的论文、代码、数据集、工件与仓库信号。论文：有；代码：无；数据：无；命中信号：数据集

图谱声望 15%

5.7

基于可用 OpenAlex 引用关系构建局部引用图，并计算 PageRank 风格的图谱声望。局部 PageRank：0.0016

引用影响力 15%

使用 OpenAlex 形态的引用元数据作为文献关注度信号，并与论文本身质量分开处理。引用数：0

引用速度 10%

引用速度按发表年限估算年均引用，降低旧论文天然占优的偏差。年均引用：0.00

研究版图角色

前沿论文

排序敏感性

稳定性：volatile；排名波动范围：476。

关键词评分

generative world model

world model

video world model

world simulator

interactive world model

world dynamics prediction

model-based reinforcement learning world model

深度分析

创新点

SPW-Nav: a streaming panoramic world model that understands movement instructions and generates one minute of 2K 360-degree video in real time from a single panorama
Interprets each instruction in the previously generated panorama as camera motion
Spherical rotation decoupling applies rotation exactly on the sphere
Pose-aligned conditioning keeps translation inputs bounded over long streams
Multi-term memory with a few-step generator continues the scene as instructions change
SPW-NavSet dataset of panoramic videos with camera trajectories and verified instructions
Supports on-the-fly instruction switching

方法

SPW-Nav takes a single panorama and language movement instructions, interpreting each instruction in the previously generated panorama as camera motion. It uses spherical rotation decoupling to apply rotation exactly on the sphere, pose-aligned conditioning to keep translation inputs bounded over long streams, and a multi-term memory with a few-step generator for continuation as instructions change. The approach is evaluated using the newly built SPW-NavSet dataset of panoramic videos with camera trajectories and verified instructions, comparing against prior panoramic generators on camera-following accuracy and video quality.

关键结果

Language-driven SPW-Nav outperforms prior panoramic generators in camera-following accuracy and video quality, and it supports on-the-fly instruction switching while generating one minute of 2K 360-degree video in real time from a single panorama.

资源

论文