WorldOlympiad: Can Your World Model Survive a Triathlon?

2026-06-09

AI Summary

Existing video generation world models lack systematic evaluation of physical plausibility, 3D structural consistency, and long-horizon controllable interaction. This work proposes the first benchmark that decouples world model assessment into three interpretable dimensions—physical fidelity, geometric consistency, and interaction faithfulness—spanning diverse applications including gaming, robotics, and real-world scenarios. The benchmark enables comprehensive diagnostics through techniques such as object segmentation, multimodal large language model–based discrimination, Gaussian splatting reconstruction, cross-view consistency analysis, and segmented action alignment. Experiments across multiple state-of-the-art models reveal significant deficiencies in their structural reasoning capabilities, underscoring the necessity and efficacy of structured evaluation frameworks for advancing world modeling.

Abstract

We introduce WorldOlympiad, a benchmark for diagnosing video-based world models across physical faithfulness, geometric consistency, and interaction fidelity. While existing benchmarks often focus on visual quality, semantic alignment, or short-term temporal coherence, they provide limited insight into whether generated videos obey physical rules, preserve coherent 3D structure, and sustain controllable interactions over long horizons. To address this gap, WorldOlympiad decomposes world-model evaluation into three complementary dimensions. The physical track uses object segmentation and MLLM-as-judge to assess whether generated videos follow interpretable rules in mechanics, thermal phenomena, and material properties. The geometry track reconstructs generated videos with Gaussian splatting and evaluates structural consistency, cross-view coherence, and camera-trajectory alignment. The interaction track assesses whether generated rollouts follow complex action prompts and maintain smooth, coherent transitions across consecutive video chunks. WorldOlympiad further covers three major downstream scenarios, including gaming, robotics, and general real-world videos, capturing diverse challenges from interactive control and embodied manipulation to open-domain motion and camera dynamics. Together, these tracks and scenarios form a scalable and interpretable evaluation suite that exposes failure modes beyond generic video quality. Experiments on state-of-the-art models reveal substantial gaps in physical reasoning, 3D consistency, and long-horizon interaction, underscoring the need for more structured evaluation protocols for generative world models.

Problem

• world models
• physical faithfulness
• geometric consistency
• interaction fidelity
• video generation

Innovation

• world models
• physical faithfulness
• geometric consistency
• interaction fidelity
• video generation benchmark