Open-source world model learns to predict Rocket League gameplay

A researcher trained MIRA, a multimodal world model that predicts future game states in Rocket League by learning from raw video and actions. The project demonstrates how foundation models can simulate complex, physics-driven environments—potentially useful for game dev tools, AI training, or anyone building interactive simulations without hand-coded physics engines.