Reactive Agents for Game AI: An ABL-Style Approach as an Alternative to Traditional NPC Architectures*
Session Number
2
Advisor(s)
Ian Horswill, Northwestern University
Discipline
Computer Science
Start Date
15-4-2026 11:10 AM
End Date
15-4-2026 11:55 AM
Abstract
Traditional game AI architectures, including Finite State Machines (FSMs) and behavior trees, face significant scalability limitations as game environments grow more complex, requiring developers to manually script each behavioral state and transition. Goal-Oriented Action Planning (GOAP) introduced declarative planning to games but brought computational overhead and complex authoring requirements. This project proposes an ABL-style (A Behavior Language) reactive planning system as a more practical alternative, implementing continuous precondition evaluation and automatic replanning without the full complexity of multi-agent coordination frameworks such as IPEM or EMA. The system is evaluated within an asymmetric turn-based strategy game inspired by Pandemic and Plague Inc., in which player actions evolve a plague that AI agents must dynamically contain and cure. By decoupling goals from actions and enabling runtime replanning, the architecture is expected to demonstrate superior adaptability and modularity compared to FSMs, behavior trees, and GOAP, while remaining accessible for practical game development. The project aims to assess whether streamlined reactive planning alone constitutes a meaningful advancement for contemporary game AI systems.
Reactive Agents for Game AI: An ABL-Style Approach as an Alternative to Traditional NPC Architectures*
Traditional game AI architectures, including Finite State Machines (FSMs) and behavior trees, face significant scalability limitations as game environments grow more complex, requiring developers to manually script each behavioral state and transition. Goal-Oriented Action Planning (GOAP) introduced declarative planning to games but brought computational overhead and complex authoring requirements. This project proposes an ABL-style (A Behavior Language) reactive planning system as a more practical alternative, implementing continuous precondition evaluation and automatic replanning without the full complexity of multi-agent coordination frameworks such as IPEM or EMA. The system is evaluated within an asymmetric turn-based strategy game inspired by Pandemic and Plague Inc., in which player actions evolve a plague that AI agents must dynamically contain and cure. By decoupling goals from actions and enabling runtime replanning, the architecture is expected to demonstrate superior adaptability and modularity compared to FSMs, behavior trees, and GOAP, while remaining accessible for practical game development. The project aims to assess whether streamlined reactive planning alone constitutes a meaningful advancement for contemporary game AI systems.