Multiplayer Interactive World Models with Representation Autoencoders

Research official 2 src. ~1 min

This paper introduces the first world model that conditions on simultaneous action streams from multiple agents, learning to correctly attribute scene changes to individual players rather than treating co-players as background. Trained on 10,000 hours of Rocket League gameplay, the 5B-parameter latent diffusion model generates four-player matches in real time at 20 fps on a single Nvidia B200 GPU. Despite training on short clips, rollouts remain stable for over five minutes in testing.

Why it matters

236 HF Daily upvotes (July 7); extends world models from single-agent to genuinely multiplayer settings—a prerequisite for using world models to evaluate and train multi-agent policies. Dataset, training code, and inference codebase are released publicly.

Importance: 4/5

236 HF Daily upvotes (+1 for ≥100); first multiplayer interactive world model; full open-source release.

Sources