AI & Machine Learning

Sony received 28 patents covering an exceptionally broad range of AI-driven gaming systems. Several address how players communicate and are matched with one another: one system analyzes voice patterns to detect personality and mood, then pairs compatible players while filtering toxic audio in real time; another predicts how long a gaming session will last based on player history, helping teammates align on time commitments before joining a match; a third translates physical gestures and sign language into context-aware chat messages that adjust in tone and content based on what is currently happening in the game. On the content moderation side, one patent filters and rewrites in-game messages so they remain relevant to each recipient's current gameplay situation, while another uses neural networks to detect and obfuscate user-specified objectionable content in games and movies, including replacing it with deepfake material. Player guidance is another recurring theme: a ghost player system generates real-time control inputs for a visible guide character that shows players exactly how to clear a challenge, while a separate crowd-sourced help system aggregates behavioral data across players to surface which strategies and tools have the highest success rates for specific obstacles, personalized to a player's own inventory and settings. Sony also received patents for avatar creation from multiple angles, including one that generates game-ready characters from user selfies while matching each game's art style, and another that combines motion capture from standard video with natural language prompts to create fully animated custom characters. For players who prefer voice interaction, a spatial audio processing system handles simultaneous voice commands from multiple players without requiring manual input switching. Accessibility is addressed through an automated narration system that uses large language models and voice synthesis to insert audio descriptions into gaps in game dialogue, matching the emotion and timing of surrounding content without any manual scripting. The remaining patents cover a wide span: a kernel function-based system that passively infers player preferences from behavioral signals and adapts audio, video, and gameplay settings accordingly; an AI that generates personalized game music in real time based on gameplay state and player interaction history; a cross-session profiling system that customizes NPC dialogue based on how individual players have engaged with characters over time; a content modification system that adapts dialogue, visuals, and audio granularly during live gameplay rather than relying on static difficulty toggles; a biometric capture system that feeds physiological data into both spectator interfaces and NPC behavior, enabling emotional bookmarking and stress-responsive enemy AI; an emotion-sensing system for VR that adjusts virtual environments based on inferred player state without requiring explicit input; an AI game streamer that modifies its own gameplay strategy in response to real-time viewer reactions; a crowd-sourced reinforcement learning system where spectators approve or disapprove of AI moves to train game agents; a system that uses video, controller inputs, and audio to auto-generate structured game context metadata for legacy titles that lack developer APIs; a recommendation engine that draws on community data to suggest optimal items and upgrades for specific in-game challenges; a 2D vector graphics to 3D asset conversion system that uses geometric metadata to improve reconstruction quality; a system that turns viewer comments and votes on gameplay clips into labeled training data for auto-commentary AI; an AI that predicts audience engagement for game trailers before release to help optimize marketing content; a highlight collage generator that trains a personalized moment model on each user's play style and composes shareable assets automatically; an AI that uses computer vision to intelligently prioritize gameplay-critical elements when streaming from console to mobile; an LLM-powered development tool that auto-generates cross-referenced JSON schemas and JavaScript code for game objects; and finally a context-aware gesture and voice input system that uses game state data to create dynamic time windows during which commands are accepted, preventing accidental inputs during ordinary movement or conversation.

EA received 7 patents spanning character animation, player behavior, and in-game assistance. One system generates character poses procedurally in real time without referencing stored motion capture libraries, learning the patterns of human movement during training and producing contextually appropriate animations on demand. Another addresses motion capture cleanup by using machine learning to predict missing marker data automatically, removing the need for animators to correct gaps frame by frame. A third patent produces expressive facial animations from speech audio by first detecting emotional tone and then conditioning the animation model on that emotion, capturing both what a character says and how they say it. For player-driven customization, EA received a patent for a system that lets players modify character poses at runtime and uses ML to generate realistic motion sequences from those single-pose adjustments. On the player support side, one system maps player behavior types using neural network clustering and delivers personalized tutorials and guidance that update based on whether a player's behavior is shifting between categories. A geometric analysis system provides context-aware tactical guidance by processing spatial relationships between game objects in real time, rather than relying on pre-scripted hints. Rounding out EA's portfolio, a patent formalizes awareness-bounded NPC decision-making as a structured technical system, explicitly modeling and enforcing what information an NPC is permitted to act on rather than relying on ad hoc line-of-sight workarounds.

Beijing Zitiao Network Technology received 2 patents focused on human-AI cooperation in multiplayer games. One introduces a communication prediction model that allows AI agents to generate contextually appropriate messages to human teammates, expressing the AI's intentions based on game state and existing chat context. The other enables game bots to adjust their difficulty and cooperation style depending on whether they are allied with or opposing a human player, transforming their perception of the game state asymmetrically in real time.

Apple received 1 patent for a motion planning system designed for virtual characters in XR and VR applications. Rather than requiring a custom motion planner for each individual character, the system analyzes the characteristics of a character's motion controller at runtime and generates movement paths that are compatible with that controller's specific capabilities, using rapidly-exploring random trees and machine-learned root motion models to produce verified plans on the fly.

Activision Blizzard received 1 patent for a system that creates NPCs trained to mirror specific human players. By continuously ingesting granular behavioral telemetry from real players and applying machine learning models to that data, the system produces bots that replicate a particular human's skill level, decision patterns, and playstyle rather than defaulting to generic AI behavior.

Adeia received 1 patent for an esports VOD recommendation system that filters content through 3 sequential phases: pre-game expectations, in-game performance, and post-game aggregated statistics. A video is only recommended if performance data from each phase validates the prior one, eliminating recommendations driven by team popularity or ratings rather than demonstrated play quality.

ByteDance received 1 patent for a MOBA matchmaking system that clusters players by skill level across multiple parallel pools. The clustering model is trained with 2 simultaneous constraints, ensuring that skill differences between teams remain within acceptable bounds while keeping pool sizes balanced enough to minimize wait times.

Colopl received 1 patent for an NPC interaction system that generates character responses dynamically using AI rather than selecting from a library of pre-authored dialogue options. This allows NPCs to produce behaviors and statements that were never explicitly programmed, giving characters a greater sense of agency during player interactions.

Cygames received 1 patent for an automated quality assurance system that analyzes game logs to identify which specific combinations of cards or game states are responsible for bugs in complex multiplayer sessions. By statistically comparing the frequency of action-state combinations in failed versus successful sessions, the system identifies causality automatically, including cases where no single element causes a problem but a specific combination does.

Dell received 1 patent for a gaming controller that shifts its physical center of gravity in response to in-game events. Unlike controllers that rely on vibration motors alone, this mechanism communicates directional in-game physics and events by actually changing the balance of the controller in the player's hands.

GDM Holding received 1 patent for a multi-agent reinforcement learning framework that trains AI agents by having them learn from one another through a matchmaking policy. The system pairs agents using different learning strategies and hyperparameters, allowing each agent to benefit from exposure to the behaviors of others, producing a collectively improving population rather than agents trained in isolation.

Gna received 1 patent for a machine learning system that segments game players into behavioral clusters based on actual in-game actions and gameplay patterns rather than demographic data. The resulting clusters are designed for use in targeted advertising, matching game promotions to players whose behavior profiles align with a given game type.

Hangzhou Electronic Soul Network Technology received 1 patent for a proactive toxicity detection system that analyzes gameplay mechanics, performance metrics, and social data together during a single match to identify behaviors such as trolling, griefing, and intentional idling. By combining those dimensions automatically rather than waiting for player reports, the system detects problematic behavior that reactive reporting systems routinely miss.

Kawasaki Jukogyo Kabushiki Kaisha received 1 patent for a system that translates a player's game controller inputs into commands for an industrial robot using AI, without the player needing any awareness of or training in robot operation. The game content and the robot task are deliberately kept separate, so a player's successful in-game actions drive real robotic execution through an intermediary translation layer.

Koei Tecmo Games received 1 patent for a battle formation management system in 3D games that automatically repositions characters pushed out of formation during combat. The system is specifically designed to handle diagonal clashes where troops intersect at oblique angles, a scenario where leader-following mechanics previously failed to prevent unnatural mingling or gaps between units.

Microsoft received 1 patent for a game development system in which generative models convert natural language prompts from designers into narrative game content, and then continue generating and adjusting content during live gameplay based on implicit player engagement signals. This creates a 2-stage loop where AI assists designers before release and continues adapting the experience for players after it.

Nintendo received 1 patent for a system that produces wave-like crowd animation effects by having NPCs imitate player actions in progressively expanding zones, with each zone delayed by both distance and a randomized waiting period. An optional synchronization layer ties the timing of imitation across all characters to the game's background music, keeping the effect rhythmically coherent rather than chaotic.

Ovomind received 1 patent for a system that predicts how players will emotionally respond to a video game sequence before they experience it. The system combines multi-modal analysis of audio, video, and graphical content with timestamped biosignal data collected from prior sessions, building predictive emotional models that work prospectively rather than simply reacting to signals after the fact.

Riot Games received 1 patent for a skill ranking system that decomposes match outcomes into discrete in-game events, each carrying its own weight toward adjusting a player's rating. Rather than updating rankings based solely on whether a team won or lost, the system attributes individual contributions within team environments, allowing ratings to converge toward a player's true skill level more quickly.

Tencent received 1 patent for an open-world NPC management system that switches each character between behavior trees of different computational complexity depending on how close that character is to the player. NPCs near the player run richer, more detailed AI, while distant NPCs use simpler trees, reducing server load without degrading the quality of interactions players actually encounter.