Standardizing Hybrid Game AI Pipelines for Broader Reuse

Establish standardized, reusable patterns and interfaces for hybrid game AI pipelines that combine lightweight scriptable decision frameworks (e.g., behavior trees, finite state machines, and utility-based systems) with selective calls to large language model or other generative components, enabling broader reuse across engines, genres, and production tooling ecosystems.

Background

The paper notes that industrial game engines and commercial titles have adopted hybrid AI pipelines which integrate traditional, lightweight scriptable decision systems with selective generative components. While these approaches can yield emergent-like behaviors and maintain performance, they are deeply coupled to proprietary tooling and tailored to specific genres or mechanics.

Because of this fragmentation, there is no shared set of reusable patterns or interfaces that would allow such hybrid techniques to be readily transferred across projects or engines. CASCADE is presented as a coordination-centric alternative, but the broader field still lacks standardized methodologies for hybridization, motivating this open challenge.