cognitive architecture design

Current cognitive architectures lack modularity and reusable reasoning support, hindering verifiability, scalability, and interpretability. Method: This paper proposes a symbolic, modular cognitive architecture grounded in the concept of “cognitive roles,” integrating metacognitive and distributed cognition theories. It decouples high-level cognitive functions—including planning, evaluation, cross-validation, and narrative integration—into well-defined, interface-explicit modules, formally modeled to ensure rigorous specification and internal consistency. Contribution/Results: The architecture introduces a role-driven—rather than task-driven—modularization paradigm, enabling dynamic inter-module collaboration and self-reflective reasoning. It constitutes the first symbolic framework that simultaneously guarantees structural interpretability and computational realizability. The work establishes both conceptual foundations and formal machinery for next-generation assistive reasoning systems that are verifiable, extensible, and cognitively grounded.

Existing AI agent architectures are typically described along a single dimension—either execution topology or cognitive function—making it difficult to characterize their design trade-offs and failure modes. This work proposes the first two-dimensional classification framework that orthogonally integrates cognitive functions (seven types, e.g., perception, memory, reasoning) with execution topologies (six types, e.g., chain, parallel, routing), yielding a 7×6 matrix that systematically defines 28 design patterns, including 15 newly named ones. Through cross-domain validation in finance, legal reasoning, network operations, and medical triage, the study distills five empirical guidelines for pattern selection and establishes a principled, framework- and model-agnostic terminology. This significantly enhances the describability and reusability of agent architectures.

How to construct human-like autonomous artificial intelligence remains an open challenge due to the lack of a unified theoretical framework defining essential functional components and gradations of autonomy. Method: This paper proposes a three-layer functional architecture: (i) a reactive layer for environment interaction; (ii) a deliberative layer that evaluates and selects behaviors based on perception and memory; and (iii) a reflective layer enabling self-modification of behavioral principles and internal structure. It introduces a progressive autonomy taxonomy—“reactive → weakly autonomous → strongly autonomous”—and develops a paradigm-agnostic theoretical framework grounded in functional decomposition, autonomy grading theory, and cross-paradigm AI design principles. Contribution: The work formally specifies necessary functional modules for human-like autonomous intelligence, establishes the first systematic autonomy taxonomy, and delivers a scalable, general theory of autonomy. This framework informs the design of strongly autonomous artificial agents, advances foundational research toward Artificial General Intelligence, and provides structural guidance for ethical governance.

The absence of a unified theoretical framework for mind and intelligence impedes progress in cognitive science and artificial intelligence. Method: This paper introduces the *Function-Alignment Theory*, which formalizes how meaning, explanation, and analogy emerge from interactions among hierarchical representations. It proposes the novel concept of *bounded interpretability*—integrating bounded rationality, symbol grounding, and analogical reasoning—and adopts a philosophy-agnostic, structural modeling approach that bridges computational modeling, cognitive science, and cross-cultural practices (e.g., contemplative meditation). Contribution/Results: The theory constitutes the first unified framework for mind and intelligence that simultaneously satisfies intuitive plausibility and formal rigor. It provides a foundational theory and modeling paradigm for explainable AI and brain-inspired systems, advancing interdisciplinary integration across computation, cognition, and embodied practice.

Current agent system designs often lack grounding in systems theory, resulting in ad hoc architectures prone to hallucination and reasoning flaws that undermine reliability. This work addresses this gap by introducing systems theory into agent architecture design for the first time, proposing a structured framework composed of five core functional subsystems. Building on this foundation, the authors abstract twelve reusable and clearly categorized agent design patterns. Through the reconstruction and validation of representative frameworks such as ReAct, the proposed approach effectively rectifies inherent architectural deficiencies, significantly enhancing modularity, interpretability, and reliability. This contribution establishes a standardized language and a structured development paradigm for agent engineering, offering a principled foundation for future research and practice.

Existing human-robot collaborative design frameworks lack temporal coordination reasoning support for dynamic, unstructured environments. Method: This paper proposes a networked computational framework integrating functional modeling and graph-theoretic representation. It explicitly models the temporal evolution of joint tasks, environmental constraints, and coordination requirements, enabling qualitative and quantitative co-analysis of coordination strategies at the conceptual design stage for the first time. The approach combines functional modeling, graph-theoretic modeling, temporal analysis of coordination requirements, and case-driven exploration of the trade-off space using post-disaster robotics scenarios. Results: Experiments demonstrate that the framework effectively identifies critical collaborative capabilities, uncovers temporal patterns in coordination overhead, and significantly enhances systematic early-stage reasoning about human-robot cooperation requirements—overcoming limitations of traditional static or real-time frameworks in temporal coordination modeling.

This work proposes a “layered attribution” diagnostic framework to disentangle the origins of inscrutable behaviors exhibited by AI agents in complex social systems, which are often conflated between internal representations and external constraints. The framework systematically distinguishes a foundational computational layer—encompassing architecture, memory, and perception—from a behavioral modulation layer comprising identity, goals, social interactions, and institutional constraints, thereby integrating representation learning, multi-agent modeling, and institutional analysis into a unified two-tier diagnostic architecture. It yields three key insights: behavioral substitutability validity hinges on the coupling among model, task, and layer; human–AI behavioral discrepancies can serve as diagnostic signals; and effective governance presupposes precise source attribution. This approach establishes a theoretical foundation for interpreting and governing AI behavior.

Existing learning modeling approaches fragment critical constructs—such as cognitive load, comprehension evolution, and subjective evaluation—lacking a unified, scalable formal framework. Method: This paper introduces a five-layer formal description language for learning dynamics, grounded in state variables, hierarchical mappings, and separation of concerns. It implements multi-faceted co-characterization through explicit structural mechanisms. Contribution/Results: We propose the novel “hierarchical responsibility separation” architecture, explicitly decoupling load generation, comprehension transformation, observation, and evaluation. Cognitive load is redefined as an interactional quantity between internal and external factors; subjective evaluation is abstracted as a minimal regulatory interface. The framework imposes no prior assumptions on functional forms or optimization objectives. Leveraging formal syntax, structured coordinates, and multi-level modeling, it provides a theoretically rigorous yet empirically compatible foundational description layer for human learning analysis and AI-driven adaptive educational systems.

This work addresses the high latency, excessive energy consumption, and behavioral incoherence arising from the tight coupling of planning, reasoning, and execution in current AI systems. To overcome these limitations, the authors propose Tri-Spirit, a novel architecture that decouples cognitive processing into three distinct layers—planning (Super Layer), reasoning (Agent Layer), and execution (Reflex Layer)—mapped onto heterogeneous hardware and coordinated via an asynchronous message bus. The framework introduces several key innovations, including a habit compilation mechanism, a convergence-based memory model, parameterized routing strategies, and explicit safety constraints. Experimental results demonstrate that Tri-Spirit reduces average task latency by 75.6%, cuts energy consumption by 71.1%, decreases large language model invocation frequency by 30%, and enables 77.6% of tasks to be completed efficiently in offline settings.