The paper presents a multiplayer interactive world model for highly dynamic environments, using Rocket League as its testbed. Trained on 10,000 hours of gameplay from publicly available bots, the 5-billion-parameter latent diffusion model conditions on the action streams of four players and reportedly generates matches at 20 frames per second on a single Nvidia B200 GPU. The authors study representation codecs, generative objectives, multiplayer conditioning, scaling behavior, and physical-understanding evaluations. They also announce releases of the dataset, full training and inference code, and a live demo.
No heat snapshots are available in the last 24 hours.