脑认知启发的可插拔BDI心智层系统设计
Brain-inspired Pluggable BDI Cognitive Layer System Design
摘要
针对当前自主智能体系统普遍存在的决策过程不透明、认知状态不可观测、认知层与执行层紧耦合等问题，提出一种脑认知启发的可插拔BDI心智层智能系统设计——EgoMind-BDI。该系统在OpenClaw执行层之上封装显式的BDI认知模块，包含信念、愿望、意图三个核心组件。信念模块维护智能体的内部世界模型，支持确定性事实、不确定信念和语义记忆三种知识表示；愿望模块管理用户目标集合并支持运行时优先级调整；意图模块采用大语言模型动态规划器生成计划，并经技能契约适配器执行三层校验后下发执行。桥接层提供标准化表述性状态转移接口，支持对信念、愿望、意图的实时查询与干预。设计了双通道认知交互模式，通过统一消息路由器将用户输入分流为任务指令与认知指令。三个典型案例的走查验证了该系统在认知透明度、可干预性和通用自动化方面的优势。与WorkBuddy、Manus等6种市场产品及LLM+BDI本体、POMDP-BDI等5种学术路线的对比分析表明，EgoMind-BDI是唯一同时在认知-执行解耦、认知层可观测、通用场景适配和知识驱动校验四个维度实现突破的系统。该工作为构建兼具推理深度与执行可靠性的下一代智能系统提供了可复现的设计范式。
Abstract
To address the prevalent issues of opaque decision-making, unobservable cognitive states, and tight coupling between cognition and execution in current autonomous agent systems, this paper proposes a brain-inspired intelligent system design featuring a pluggable BDI cognitive layerEgoMind-BDI. The system encapsulates an explicit BDI cognitive module atop the OpenClaw execution framework, comprising three core components: a belief module maintaining the agent's internal world model with support for deterministic facts, uncertain beliefs, and semantic memory; a desire module managing user goals with runtime priority adjustment; and an intention module employing a large language model as a dynamic planner whose outputs undergo three-layer verification by a skill contract adapter before execution. The bridge layer provides standardized representational state transfer interfaces for real-time querying and intervention on beliefs, desires, and intentions. A dual-channel cognitive interaction mode is designed, routing user input into task commands and cognitive commands through a unified message router. Three case walkthroughs demonstrate the system's advantages in cognitive transparency, intervenability, and general-purpose automation. Comparative analysis with six market products including WorkBuddy and Manus, and five academic approaches including LLM+BDI Ontology and POMDP-BDI, indicates that EgoMind-BDI is the only system achieving simultaneous breakthroughs in cognition-execution decoupling, cognitive observability, general scenario adaptation, and knowledge-driven verification. This work provides a reproducible design paradigm for constructing next-generation intelligent systems that combine human-level reasoning depth with software-level execution reliability.关键词
BDI模型/脑认知启发/认知架构/全状态可观测/知识驱动规划/可解释人工智能/技能契约/分层解耦Key words
BDI model/brain-inspired cognition/cognitive architecture/full-state observability/knowledge-driven planning/explainable artificial intelligence/skill contract/layered decoupling引用本文复制引用
裴震,高武奇,罗钧旻.脑认知启发的可插拔BDI心智层系统设计[EB/OL].(2026-05-21)[2026-05-22].https://chinaxiv.org/abs/202605.00148.学科分类
计算技术、计算机技术
评论