AI & Machine Learning

Looking at EA's 3 patents, the company received technology aimed at streamlining animation workflows and improving NPC behavior. One patent uses machine learning to automatically predict missing motion capture marker data, eliminating the need for frame-by-frame manual corrections. Another lets players modify character animations during active gameplay, with AI generating realistic motion variations from single pose adjustments without designer intervention. The third creates a structured framework where NPC knowledge is explicitly modeled and constrained to match what the player actually knows, moving beyond simple line-of-sight checks to prevent frustrating all-seeing enemy behavior.

Sony received 12 patents covering a wide range of AI-driven game and media technologies. Several address content adaptation and personalization: one system modifies in-game dialogue, visuals, and audio in real-time based on player behavior and accessibility needs, while another uses kernel functions to detect player preferences through behavioral signals and auto-adjusts settings across audio, video, and gameplay dimensions. The company also patented an AI ghost player that generates contextual guidance by analyzing live game state and producing control inputs for a visible guide character, distinct from static hint systems. Two patents create crowd-sourced training pipelines where viewer votes and comments on gameplay clips generate labeled data for AI commentators, and spectator feedback during streams trains game bots through human approval signals. Additional patents include an emotion-sensing system that adjusts virtual environments based on inferred physical state without manual input, an AI-powered content filter that detects and obfuscates objectionable material using deepfake replacement, and voice analysis technology for matchmaking that pairs players by personality and mood compatibility while filtering toxic audio. Development tools include an LLM system that auto-generates cross-referenced JSON schemas and JavaScript code for game objects, and another that converts 2D vector graphics into 3D assets by using geometric metadata alongside raster data. Sony also patented technology that auto-generates shareable highlight collages by training a personalized model on user play style, a streaming system that uses computer vision to detect and enhance gameplay-critical assets when moving from console to mobile, and matchmaking that analyzes voice patterns for social compatibility.

Adeia's AI-powered system recommends esports and gaming video-on-demand content using a sequential multi-phase filter that validates pre-game expectations against actual in-game performance and post-game statistics, moving beyond traditional ratings or watch history approaches.

Beijing Zitiao Network Technology received 2 patents focused on human-AI cooperation in multiplayer games. One enables AI agents to communicate battle intentions to human teammates through a communication prediction model that synthesizes game state, existing chat context, and inferred intent to generate contextually appropriate messages. The other adjusts AI difficulty and cooperation style based on team alignment, transforming how the bot perceives game state differently depending on whether it's allied with or opposing the player.

Koei Tecmo Games patented an AI system that maintains orderly battle formations in 3D games by automatically repositioning attacked characters back toward their own troops, specifically solving the oblique-clash problem where diagonal troop intersections create unnatural mingling that leader-following mechanics fail to prevent.

Activision Blizzard patented AI-powered NPCs that learn from real players using neural networks, creating bots that mimic specific human behavior, skill levels, and playstyles by continuously ingesting granular behavioral telemetry, enabling impersonation of friends or professional players across multiple platforms and genres.

Riot Games patented a dynamic skill ranking system that adjusts player ratings based on discrete in-game events rather than binary win/loss outcomes, with each weighted event contributing independently to enable accurate individual ranking within team environments and faster convergence to true skill levels.

Kawasaki Jukogyo Kabushiki Kaisha patented a system where gamers unknowingly control industrial robots through gameplay, using AI to translate game controller inputs into robot commands so players need no awareness of robotic operation while their successful game actions drive skilled robot execution.

Tencent patented a system that dynamically switches NPC behavior trees based on player proximity, using distance as the primary signal to vary computational complexity with nearby NPCs receiving rich behavior trees and distant ones running simplified versions, enabling scalable open-world populations without degrading perceived quality.