AI & Machine Learning

Sony received 13 patents addressing different aspects of AI in gaming. One application describes a system where players can delegate gameplay to AI agents with granular permission controls, whitelisting or blacklisting specific in-game actions rather than handing over full control. Another lets gamers search gameplay videos by describing scenarios and controller inputs, with AI automatically indexing videos using game-specific metadata like input speed and completion times instead of manual tags. A third patent covers a platform-level AI engine that transforms game visuals and audio in real-time to match accessibility needs without developer intervention, identifying and adapting even occluded elements while learning from player feedback. Sony also filed a patent for analyzing aggregated gameplay data to identify which activities are statistically most likely to unlock specific trophies and rare items, quantifying the relationship without relying on manual guides. Eye-tracking technology appears in another application, where gaze data drives real-time adjustments to audio, visuals, and immersion levels based on where players look and how engaged they appear. A related accessibility patent describes AI that detects physical limitations from controller input patterns and automatically compensates in real-time, filling in actions the player cannot physically perform. The matchmaking application combines implicit behavioral data with explicit preferences and uses machine learning to pair players based on complementary rather than purely similar skill levels. Resource allocation gets addressed through a patent that predicts player skill from behavior and pre-allocates computing power accordingly, shifting from hardware-driven to behavior-driven optimization. Another application clones real players as AI companions by extracting behavioral attributes from recorded gameplay, letting solo players team up with NPCs that mirror actual human playstyles. Sony also filed for a system that analyzes console gaming history to auto-generate personalized trailers by dynamically selecting content sections tailored to individual users. A coaching-focused patent watches gameplay and controller inputs to generate real-time instructions or custom practice minigames targeting specific opponent weaknesses. Cross-platform streaming gets covered by a patent that enhances visibility of critical game assets when moving from console to mobile, using AI to understand which elements matter and selectively boosting them rather than blindly resizing frames. The final application combines motion-captured body movements with voice descriptions to generate both avatar appearance and matching animations simultaneously through generative AI.

Nvidia received 4 patents spanning visual customization, video synthesis, animation, and quality assurance. The visual customization application translates natural language descriptions into executable shader code in real-time, letting players describe desired looks and instantly generate corresponding visual adjustments without preset menus. A video generation patent separates content appearance and motion into independent controllable latent spaces within a generative adversarial framework, allowing creators to vary visual style and movement dynamics independently. The animation training application combines motion tracking rewards with distribution matching in a single reinforcement learning framework, producing character movement that is both realistic and adaptable to new situations. Automated bug detection appears in the fourth patent, where vision-language models analyze gameplay video and audio to identify issues without pre-defined rule sets, using embedding dissimilarity scores to select keyframes representing meaningful state changes.

EA filed 3 patents covering NPC perception, level generation, and character animation. The NPC application structures AI around sensory perspective, mapping what characters can actually see, hear, or detect through distinct detection models per sense type rather than letting them act on information they could not realistically perceive. A level generator uses encoder-decoder neural networks conditioned on structured features like difficulty, size, and symmetry, letting designers specify constraints and receive levels that satisfy those properties rather than accepting uncontrolled generative output. The animation system employs a variational autoencoder that generates character poses from learned distributions and environmental context at runtime, eliminating the need to store or retrieve motion capture libraries during gameplay.

Skydance Silicon Valley received 2 patents for automated camera management in games. One application uses a neural network to dynamically evaluate and select camera angles in real-time, scoring viewpoints and cutting between them with threshold gating to prevent erratic switching while maintaining responsiveness. The other predicts per-player reorientation time when switching camera control schemes and converts player hints into weights that subtly influence automated camera selection without breaking immersion or requiring direct control.

Hasbro filed 1 patent for an AI referee system in trivia games that positions independent validation models as a governance layer between the user interface and generative AI, using pre-loaded query context to catch hallucinations and rule violations before they affect gameplay rather than relying on single-model self-correction.

SafeBear received 1 patent for a multi-brain AI system that detects cyberbullying in social networks and video games by deploying parallel AI models, each specializing in a distinct toxicity theme, then masks harmful content in real-time while recording evidence on blockchain for potential legal proceedings.

Rpg Fun filed 1 patent for an AI-powered game engine that acts as an infinite game master, generating worlds, narratives, and gameplay in real-time from player interactions using Associative Retrieval Augmented Generation for direct game state lookups and a dual-oracle architecture separating low-latency responses from deeper world-generation tasks.

Bitpart AI received 1 patent for a multi-agent planning system that converts human-authored narrative transcripts into hierarchical task networks of re-composable plans, letting a runtime director dynamically select and sequence NPC behaviors across multiple agents rather than relying on fixed scripts or single-character AI.

Adeia filed 1 patent for a gameplay coaching system that imports skilled players' AI-trained behavioral models into live sessions, providing contextual real-time advice that adapts as the player deviates from recommended actions, with optional generative AI delivering guidance in the skilled player's actual voice.

WMC Analytics received 1 patent for a replay analyzer that parses game files, classifies events, and generates training assessments identifying the least-lacking skill relative to a player's goal, prioritizing coaching on the highest-leverage improvement rather than the weakest absolute skill through path-of-least-resistance calculations.

Unity Technologies SF filed 1 patent for on-device machine learning that profiles users from gameplay behavior inside the game engine runtime, transmitting only abstracted interest profiles rather than raw behavioral data to enable privacy-safe ad targeting and recommendations without exposing personal information.

Microsoft received 1 patent for an AI-powered creation system where GPT models assist designers in building narratives during development while player behavior triggers generative content during live gameplay, with implicit engagement metrics feeding back into iterative game versioning in a closed-loop system.

SoftBank Group filed 1 patent for a mobile system that analyzes real-time sports imagery to drive a personalized virtual commentator, syncing character animation with indexed motion data from multiple sports while adjusting dialogue style based on detected viewer emotions.