Weekly Patent Digest

Sony's 11 patents this week span personalization, adaptive rendering, and AI-driven gameplay in ways that often build on one another. A recommendation system learns from whether players actually act on its suggestions, retraining itself over time to improve. A separate but related filing scores games and media by cognitive load, using gameplay metrics to suggest what to play and for how long. On the social side, 1 patent proposes using a player's own in-game voice to build a socialization profile that guides matchmaking and filters or rewrites what teammates say. For cloud rendering, Sony describes a method that generates lightweight 3D placeholders from virtual camera captures and passes them to the client via metadata, reducing what the client must process locally. Another rendering patent watches player input and live game state to infer how an asset is being used, then streams it at a fidelity level matched to that use. Rounding out the AI work, 1 filing describes an AI teammate that observes how you play and quietly adjusts its own behavior to complement your style, and another trains a model on recorded NPC encounters to modify NPC appearance, behavior, or emotion either during a live session or in a future build.

EA filed 2 patents this week covering opposite ends of the multiplayer experience. One describes a sports game that scales its field geometry and player assignments dynamically based on how many people are actually in the session, with disconnected players handed off to limited CPU control and remaining players shown quick contextual prompts. The other addresses spectating, describing a system that streams simulated game state packets to a client neural renderer, which allows visual modifications and multiple viewpoints without requiring the client to hold a full copy of the game's assets.

Nintendo's 4 filings this week are focused on the edges of the play experience: getting into a session, communicating during one, and navigating within a game world. 1 patent describes a single button that either opens a communication group's chat or surfaces group-exclusive multiplayer invitations, with a host-and-guest sharing model where only the host actually runs the game. A second filing outlines a pre-game camera setup that places multiple players into labeled on-screen zones and confirms each face before the game begins. A third describes a companion-device map system where players select destinations in 2D and receive voice or display guidance toward the corresponding 3D in-game positions, with layered fields and arrival-triggered audio. The fourth covers headset disconnection handling on a controller, restricting audio output and restoring it when the player reconnects or provides input, with a speaker fallback option.

Apple and Meta each filed 1 patent in the VR and AR space. Apple's filing describes a deep learning system combining causal reprojection with gaze prediction to generate frames and apply foveated rendering on wearable headsets, reducing processing demands during extended use. Meta's patent takes a different angle, describing a process for deriving consistent avatar design constraints by randomly generating test avatars, measuring facial feature proportions, and encoding those measurements as rules that apply across social, XR, and game platforms.

Microsoft and PostVoyage both filed 1 patent each touching NPC behavior, though from very different directions. Microsoft's filing describes an AI decision engine that restricts NPC choices to a developer-curated action list, drawing on character memory and world state to guide decisions rather than generating open-ended dialogue. PostVoyage's approach is considerably more open-ended: an LLM runs inside the game engine at runtime, reasons over per-NPC alignment vectors and interaction history, and rewrites world state so that factions, quests, and items shift across sessions based on what has happened before.

Nvidia filed 1 patent describing a pipeline for building computer vision training datasets from game footage without manual labeling. The method crawls game wikis, uses a multimodal LLM to label a seed set of images, and then expands that set automatically from gameplay video.

Niantic filed 1 patent for an AR localization approach that trains a single scene-agnostic coordinate model paired with a lightweight per-scene map code, allowing a game server to place AR content in a new location without constructing a 3D model or retraining from scratch for each site.

Pixelworks Semiconductor Technology filed 1 patent describing a motion vector filtering method that operates along the trajectory of motion to reduce interpolation holes when generating in-between frames, targeting smoother framerates in real-time 3D rendering on handheld and mobile devices.

Stern Pinball's 1 filing this week describes pinball machines that save a player's session state to a remote server by user ID and restore it in a later visit, turning what is normally a self-contained game into a system that carries progression across sessions.