Gamma-World: Generative Multi-Agent World Modeling Beyond Two Players
Fangfu Liu, Kai He, Tianchang Shen, Tianshi Cao, Sanja Fidler, Yueqi Duan, Jun Gao, Igor Gilitschenski, Zian Wang, Xuanchi Ren
arXiv:2605.28816 · 431▲ · cs.CV
View on arXiv →Premium readers get an interactive explainer for this paper — a figure you can poke at, not just read.
What it is
Gamma-World is a video world model that generates synchronized, action-controllable video streams for multiple agents (game players or robot arms) sharing one environment. It introduces two components: Simplex Rotary Agent Encoding, a parameter-free extension of 3D RoPE that assigns each agent a rotary phase at the vertex of a regular simplex so agents stay distinct but permutation-symmetric, and Sparse Hub Attention, where learnable hub tokens mediate cross-agent communication instead of dense all-to-all attention.
Why it matters
Most interactive video world models are single-agent and either compress multiple views into one canvas or use dense joint attention that scales quadratically with player count and ties the model to a fixed roster. This design cuts cross-agent attention from quadratic to linear in the number of agents and, because agent identity is not a learned per-slot embedding, a model trained on two players can run four players with no retraining, which matters for real-time multiplayer or bimanual robot simulation.
Practical takeaway
If you are building multiplayer game generation or coordinated multi-robot simulators, you can now add agents without a fixed learned slot ordering or retraining, and route their interaction through a small set of hub tokens rather than paying dense pairwise attention. Watch for these RoPE-based identity tricks and hub-token attention showing up in interactive world-model tooling.
Key result
On synchronized two-agent Minecraft trajectories, Gamma-World reports FVD 184.1 / FID 24.8 on the Memory protocol versus Solaris at 333.8 / 51.7 and a frame-concat baseline at 450.6 / 69.8, with similar gains across Grounding, Movement, Building, and Consistency. Sparse Hub Attention cuts self-attention latency at 8 players to 4.5ms vs 17.6ms for dense attention. Caveat: evaluation is limited to Minecraft-style scenes plus a small real-world robot-arm demo, all against just two baselines.
Subscribe
Get the next issue.
Free. One email a week. Unsubscribe any time: no account, no dark patterns.