EA

EA's single game engine patent describes a decoupled architecture where simulation and rendering operate independently through state streams. The presentation engine can generate frames by interpolating between two state packages from the simulation without requiring additional simulation input, which breaks the traditional lock between simulation tick rate and frame rate. The system also supports multiple simulation engines running at different cycle rates, all feeding the same renderer, which allows complex games with distinct environmental zones or gameplay layers to process at speeds appropriate to their individual needs without forcing everything onto a single timeline.

Three AI and machine learning patents tackle different aspects of game content generation and character behavior. The first introduces a perception system where NPCs make decisions based only on what their simulated senses can detect, processing visual, auditory, and other sensory input through detection models that filter out information the character wouldn't realistically know about. The second applies encoder-decoder neural networks to level design, training on existing maps and allowing designers to specify constraints like difficulty, size, and symmetry to generate new layouts that meet those requirements. The third moves character animation from retrieval-based blending to real-time generative output, using a variational autoencoder that produces poses from latent variables and environmental context without needing access to motion capture libraries during gameplay.