Multiplayer world models struggle to keep shared virtual worlds consistent
MultiWorldBench: Do Independently Controlled Views Describe One Shared World?
Artificial IntelligenceComputer Vision and Pattern Recognition
Summary
When multiple players each control their own view of a virtual world, it’s hard to keep all views consistent and reflecting the same shared environment. The authors created MultiWorldBench, a Minecraft-based test with many tasks to check if different views stay synced, consistent, and persistent. They tested several multiplayer world models and found that while some handle parts well, none fully maintain a consistent, shared world. Human judgments matched these findings closely, showing current models aren’t yet good enough to create truly coherent multiplayer worlds.
What this means in practice
- •For game developers: Improve multiplayer game engines by diagnosing and addressing consistency issues between players’ views to build more reliable shared worlds.
- •For virtual environment designers: Use MultiWorldBench as a test suite to evaluate if independently controlled perspectives remain coherent in shared simulations or training systems.
Authors
Zhangbo Xu, Ruoxi Zhang, Rui Hu, Yisong Wang
Abstract
Multiplayer world models must ensure that independently controlled views remain consistent with one shared and persistent world. We introduce MultiWorldBench, a diagnostic Minecraft benchmark containing 495 case configurations across seven task suites and ten capabilities, including independent control, cross-view motion, shared-state synchronization, persistence, structural reasoning, concurrent interaction, and delayed revisit. We evaluate Solaris, Gamma-World, and MineWorld, using Engine GT as a reference. Gamma-World achieves the highest ten-capability average among the generated systems at 21.39, followed by Solaris at 20.88 and MineWorld at 1.89, while Engine GT reaches 91.69. Gamma-World performs better on several control, shared-state, and revisit capabilities, whereas Solaris leads in cross-view motion and race-condition consistency. Nevertheless, all generated systems score at most 8.00 on state persistence and 1.33 on structural consistency, and none succeeds in spatial reasoning or building-identity preservation. Human preferences produce the same overall ranking and show strong alignment with the automatic evaluation, with a mean dimension-level Spearman correlation of 0.96. These results show that plausible individual views do not yet constitute a coherent multiplayer world.