CEAA: A Cognitive Embodied Agents Architecture for... | AI Research

The Cognitive Embodied Agents Architecture (CEAA) provides a modular framework for building the "brains" of Intelligent Virtual Agents (IVAs) in 3D interactive environments. The authors, Aimilios Hadjiliasi and Louis Nisiotis, developed this architecture to bridge the gap between high-level cognitive reasoning models and the real-time performance requirements of modern game engines like Unity and Unreal Engine.

Addressing the Architecture Gap

Current approaches to virtual agents often force a choice between two extremes. Reactive systems—such as state machines or behavior trees—are efficient and easy to implement in game engines but lack the depth for long-term goal-directed reasoning or explainability. Conversely, established cognitive models like Belief-Desire-Intention (BDI), SOAR, or ACT-R provide strong theoretical foundations for reasoning but are often too complex to integrate into real-time 3D simulations. CEAA aims to solve this by providing a reusable, implementation-oriented template that combines these levels of abstraction.

How the Architecture Works

CEAA organizes agent functionality into three distinct layers: the User and Environment Layer, the Knowledge Layer, and the Agent Layer.
The architecture utilizes a "blackboard" paradigm in the Knowledge Layer, where environmental events are recorded as shared problems. Agents use a "Sense" component to selectively monitor this blackboard for information relevant to their specific goals. Once an event is identified, the agent’s "Think" component orchestrates a process involving:

• Memory: Stores past experiences and semantic knowledge to allow for context-aware behavior.

• Cognitive Constructs: Maintains the agent's internal state through Beliefs (understanding of the world), Desires (motivations and goals), and Intentions (committed plans).

• Planning and Action: Translates abstract decisions into executable behaviors that the game engine can render.

Modular Design for Real-Time Interaction

By separating the agent’s cognitive state from the physical environment, CEAA allows developers to integrate complex reasoning without tightly coupling it to the game engine’s physics or rendering logic. The architecture uses a modular approach with twelve interconnected components, such as a Memory Processor for experience management and a Behavior Mapper for grounding abstract decisions into physical actions. This structure is intended to support scalability and adaptability, allowing agents to operate autonomously while maintaining a clear, explainable decision-making process.

Franklin Analysis

The CEAA framework is designed to address the practical difficulty of deploying cognitive agents in complex virtual worlds. The authors note that while existing cognitive architectures offer rich theoretical models, they often lack the implementation guidance necessary for real-time 3D environments. By building on the Sense-Think-Act paradigm and integrating the BDI model with a blackboard-based knowledge system, the architecture provides a structured path for developers to move beyond simple reactive behaviors toward agents capable of persistent, goal-oriented interaction. The effectiveness of this approach relies on the separation of concerns between the environment, knowledge representation, and agent reasoning.