EA

The single AI and machine learning patent describes a system in which an imitation learning agent plays through a game level, generates state and action data, and uses that information to place assets and reshape terrain. The same agent then replays the modified level to check whether goals remain reachable, closing a loop that iterates between content generation, testing, and correction. Instead of relying on manual designer adjustments or fixed rule sets, the approach derives level layout directly from predicted player behavior in each state.

Graphics work spans 3 patents that tackle rendering challenges through neural networks and per-pixel techniques. One applies distance-based color blending to portions of mesh particles, assigning a first color, a second color, or a blend depending on each pixel's proximity to a defined point on a moving entity. Another replaces traditional cloth physics simulation by training a neural network to predict both compressed mesh geometry and compressed texture representations from character pose, then selecting pre-computed normal maps via nearest-neighbor lookup at runtime. The third enables game spectating by Streaming simulation state packets to a client that runs a neural renderer locally, allowing visual modifications and multiple viewpoints without requiring full game assets or high-bandwidth video streams.

A single game mechanics patent handles dynamic field scaling and player reassignment in sports titles. The system adjusts field dimensions based on active player count, with length and width shrinking at different rates, and continuously reassigns every character model to a zone by evaluating live positions, nearby objects, and control type. When a player disconnects, the system hands that character to limited CPU control and displays filtered quick prompts to remaining players, folding disconnect handling and contextual overlays into the same state-driven assignment loop rather than treating them as separate subsystems.