AI & Machine Learning

Sony received 22 patents covering AI-driven gaming experiences across content creation, player engagement, and automation. Several filings address keeping players connected to games, including a system that generates personalized story recaps when players return after breaks, another that delivers real-world notifications through in-game NPCs to maintain immersion, and a third that analyzes why players quit to create tailored tutorials addressing specific skill gaps. The company filed multiple patents on AI-assisted content generation, including one where designer keyframes act as guardrails for generative models creating game scenes, another that auto-generates gameplay recaps and tutorials by repurposing existing content, and a system that transforms session data into storybooks, podcasts, and highlight reels. Sony also filed on adaptive difficulty that learns play styles and pushes players beyond their comfort zones, quality assurance automation that detects and fixes performance issues in real-time, and controller-free gaming through multimodal sensor fusion. Additional filings cover NPC dialogue systems that listen to player conversations and generate contextually aware responses using LLMs trained on real chat data, a text-to-character system that generates custom avatars from natural language descriptions, and a machine learning approach that reverse-engineers controller inputs from gameplay video. The company also filed on collapsed gameplay that strips games down to essential story moments for time-pressed players, AI-powered haptics that convert text into tactile feedback, route-based storytelling for car trips using navigation data and LLMs, and crowd-controlled gaming where thousands of stadium fans collectively control games through body movements. Other patents include a system that analyzes gameplay clips and social comments to generate quality reports for developers, dynamic gameplay assistance in complex virtual environments, and training methods for single game agents with tunable behavior styles via sliders.

Bandai Namco received 1 patent on an avatar creation system that analyzes user behavioral data and accepts multi-modal inputs like text, image, and voice to automatically generate personalized characters for virtual spaces, eliminating manual customization while producing multiple character variants for user selection.

Lemon filed 2 patents on AI-powered game creation tools. One uses diffusion models to generate visually consistent game assets from a single prompt by creating stylistically unified prompts for all elements before image generation, ensuring characters, backgrounds, and environments share a common visual language. The other combines LLMs with diffusion models in a dual-agent architecture that lets users build games through a chat interface, iteratively adjusting both code and visual assets in real-time.

GDM Holding filed 2 patents on machine learning systems for gaming. One describes a companion agent that watches gameplay via video stream and answers questions in real-time by overlaying visual hints on the screen, working externally without requiring developer integration into the game itself. The other addresses multi-agent reinforcement learning matchmaking, where each learner has its own distinct distribution of training partners, enabling structured diversity in interactions that prevents collapse to a single dominant strategy.

Google received 2 patents on AI-powered player assistance. One system detects when players are stuck and automatically finds or creates video game tutorials in real-time, switching between play, search, and content creation modes to fill gaps by generating missing tutorials from the player's own captured footage. The other describes an AI coach that processes gameplay video and explains optimal moves through interactive chain-of-thought reasoning in natural language conversations, allowing players to interrogate the AI's strategic thinking rather than receiving opaque recommendations.

Truist Bank filed 1 patent on a player re-engagement system that combines in-game behavioral data with real-world financial transaction data from a partner banking application to train prediction models that identify and prevent player churn.

NetEase received 1 patent on an offensive zone selection system for virtual sports games that automatically matches players to optimal attack positions, shifting from passive attribute buffs to active zone-based opportunity selection through a structured algorithm that bridges player selection and spatial positioning.

EA filed 3 patents addressing player experience and accessibility. One captures real athletes' unique movement signatures and injects them into game characters by creating lightweight difference models that store only idiosyncratic deviations from standard animations, reducing storage requirements while enabling application to athletes without direct capture data. Another generates real-time contextual audio descriptions of gameplay for visually impaired players, processing game state data to describe character movements, object distances, and environmental context. The third uses geometric feature extraction from virtual environments to determine when and how to assist players navigating obstacles, enabling contextually aware help based on spatial challenges rather than performance metrics alone.

Binary Kinetics filed 1 patent on a stateless AI Game Master architecture that applies event sourcing principles to LLM context window constraints, reconstructing prompt-compatible semantic context from compact action logs rather than replaying full conversation history, combined with multi-provider LLM failover for scalable AI-driven interactive fiction.

Meaning Machine received 1 patent on an NPC dialogue system that dynamically updates world lore in real-time as characters improvise dialogue, using semantic vector comparison to merge emergent NPC-generated content back into the master context database for consistency across multiplayer sessions without exceeding LLM context limits.

Microsoft filed 4 patents on AI systems for realistic virtual character representation. One replaces classical inverse-kinematics solvers with a neural network that understands hand biomechanics and object interactions for real-time hand pose generation in games and VR, also serving as synthetic training data for AI models. Another integrates a statistical prior model trained on human appearance data to constrain Gaussian splatting-based avatar reconstruction, improving photorealism from sparse input. Two patents address voice transformation in multiplayer role-playing games, with one adjusting vocal output based on in-game character states like breathing heavily when running or sounding muffled underwater, and another using computer vision to analyze character physical attributes and automatically blend appropriate voice characteristics that match appearance while preserving the speaker's natural speech patterns.

Playtika filed 1 patent on a dynamic player segmentation system that personalizes gaming experiences in real-time based on behavioral data through a flexible infrastructure decoupled from game architecture, enabling operators to implement temporary or permanent experience modifications without deep code changes.

Roblox received 1 patent on using AI diffusion models to auto-generate customized video game character avatars from text prompts, replacing manual template-based character customization with open-ended natural language character generation.

Nintendo filed 1 patent on a smart NPC companion system that automatically detects when a player character boards a moving platform and triggers NPC repositioning and state-readiness without manual intervention, solving the problem of companion desynchronization during dynamic traversal.

Adeia received 1 patent on a time-aware deep learning framework that fits gaming sessions into user-defined time windows while maintaining consistent difficulty by embedding time constraints directly into Monte Carlo tree search algorithms via dual policy and value networks that optimize simultaneously for game outcome and session duration.