MASS: Multiplayer World Models with Authoritative Shared State
MASS separates shared world state from per-player rendering to keep multiplayer simulations consistent at scale.
The paper proposes a learned Logic Engine that updates one global typed state from all players’ actions, with no hand-written transition function. A separate Rendering Engine then produces camera-specific views from that shared state on demand. On a multiplayer Snake benchmark, the authors report better state accuracy and lower cross-view inconsistency than matched multi-view baselines. They also say MASS can roll out predicted worlds with 1,024 concurrent players for 10,000 recurrent steps. HF Daily Papers' note
The paper proposes a learned Logic Engine that updates one global typed state from all players’ actions, with no hand-written transition function. A separate Rendering Engine then produces camera-specific views from that shared state on demand. On a multiplayer Snake benchmark, the authors report better state accuracy and lower cross-view inconsistency than matched multi-view baselines. They also say MASS can roll out predicted worlds with 1,024 concurrent players for 10,000 recurrent steps. HF Daily Papers' note
score 5