MASS: Multiplayer World Models with Authoritative Shared State

2026-08-06

AI Summary

This work addresses the limitations of existing video world models in multi-agent scenarios, where coupling world states with viewpoint-dependent visual latent variables leads to computational redundancy, inconsistent viewpoints, and poor scalability. Inspired by multiplayer game architectures, the paper proposes the first explicit, authoritative shared-state mechanism that fully decouples world dynamics from view rendering. A differentiable, end-to-end learnable logic engine maintains a global, typed state representation, while a separate rendering engine generates consistent, agent-specific views on demand. The approach eliminates the need for handcrafted state-transition functions and significantly outperforms current models on a multi-agent Snake benchmark, achieving higher state prediction accuracy, markedly reduced cross-view inconsistency, and enabling long-horizon, coherent simulations with up to 1,024 agents over 10,000 timesteps.

Abstract

Current video world models struggle in multiplayer environments because they entangle world state with view-dependent visual latents, leading to redundant compute, view inconsistencies, and poor scalability. We propose MAS (Multiplayer world models with Authoritative Shared State) to resolve this limitation. Inspired by multiplayer game architectures, MAS disentangles world dynamics and view rendering. A learned Logic Engine advances a global, authoritative typed state from joint actions without any hand-written transition function, acting as the sole recurrent memory and synchronization reference. From this shared state, a learned Rendering Engine generates independent and consistent views for any requested camera on demand. This explicit disentangling allows MAS to achieve superior state accuracy and lower cross-view inconsistency compared to state-of-the-art multi-view baselines on a matched multiplayer Snake benchmark. It advances predicted worlds with 1,024 concurrent players for 10,000 recurrent steps. Our results show that explicit, authoritative state modeling provides a practical foundation for scalable and consistent multi-agent world simulation.

Problem

• multiplayer environments
• world models
• view inconsistency
• scalability
• shared state

Innovation

• authoritative shared state
• world model
• multiplayer simulation
• disentangled dynamics and rendering
• learned logic engine