AMD

1 AI & Machine Learning patent covers an approach to training non-player characters through a hierarchical structure, where more experienced "leader" NPCs guide "follower" NPCs by demonstration rather than requiring each NPC to be trained independently from the ground up. Because the leader acts as a teacher, the system avoids duplicating the full training process across every NPC in a scene, which reduces the computational load while still allowing NPCs to exhibit varied behaviors across different skill levels and roles.

The 1 Graphics patent addresses one of the more demanding aspects of real-time ray tracing: keeping bounding volume hierarchy (BVH) structures up to date as objects move within a scene. AMD's application separates the BVH's tree topology from the underlying geometry data, so the structural relationships between spatial regions can be cached and carried across frames even as object positions shift. Traditionally, rebuilding the BVH is among the most computationally expensive steps in ray tracing pipelines, and decoupling these two components reduces how often that full rebuild needs to occur.

AMD's 1 Game Engines patent describes a rendering and game engine architecture where machine learning inference runs as a core part of the pipeline, rather than as a separate layer applied after the fact. This tight integration allows the engine to adapt physics and Graphics computations in real time based on current workload demands. Rather than treating AI as an auxiliary tool bolted on to an existing system, the architecture positions inference as a fundamental component of how the engine manages and allocates its processing work.