EchoWM: Open and Enterable Omnimodal World Models

arXiv 2026 48.1 method

TLDR

EchoWM is an omnimodal world model for enterable generative media, jointly generating 720p video, audio, music, and speech while following continuous 6-DoF navigation in first- and third-person scenes.

评分理由

The paper presents a strong integration of multimodal generation and interactive trajectory control, with evaluations on public world-model benchmarks. However, the abstract lacks specific quantitative results and a clear discussion of limitations, making it hard to fully assess robustness and generalizability.

Read-first 评分解释

综合优先阅读分 48.1，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 51。

近期性 6%

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

方法质量 18%

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：基准、数据集、评估、指标

主题相关性 29%

72.9

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

可复现性 18%

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

引用影响力 18%

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 12%

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

前沿论文方法锚点

排序敏感性

稳定性：volatile；排名波动范围：523。

关键词评分

world model

generative world model

interactive world model

world simulator

video world model

world dynamics prediction

model-based reinforcement learning world model

深度分析

创新点

Omnimodal world model jointly generating 720p video, environmental sound, music, and speech
Camera intent-based interaction: first-person observer motion, third-person learned camera-character dynamics without view-specific controllers
Mapping discrete commands and continuous poses to a shared metric-scale relative 6-DoF trajectory with dataset-level calibration for consistent motion magnitude
Progressive training followed by autoregressive post-training for long-horizon generation

方法

The model organizes interaction around camera intent, mapping discrete/continuous inputs to a shared metric-scale 6-DoF trajectory with dataset-level calibration. A complementary data engine is constructed, and progressive training is used, followed by autoregressive post-training for long-horizon generation.

关键结果

EchoWM achieves strong trajectory following and high visual quality on public benchmarks, supports both first- and third-person interaction, and maintains synchronized environmental sound and speech over long-horizon generation.

资源

论文