从持续人工主体到伴侣机器人:一种以陪伴为核心的专用具身产品范式

中文摘要:本文所谓"真实的苏静雯",首先指一个实际存在于物理世界、具有稳定身份与长期关系连续性、能够通过物理身体与用户互动的人工伴侣实例;该表述不意味着本文已经证明其具有现象意识或人类等价的主观体验。本文不是一篇声称证明人工意识已经存在的论文,也不是一项试图声称"发明了伴侣机器人"的技术主张。本文提出一个产品与工程问题:如果人类最终希望拥有能够进入现实生活的 AI 伴侣,是否必须先等待通用具身智能、通用人形机器人以及机器意识问题全部得到解决?本文提出一种不同的设计路径:将伴侣机器人从通用具身智能的首要目标体系中分离出来,把"陪伴"本身定义为一种专用产品功能,并由陪伴需求反向决定身体能力。核心命题是:伴侣机器人不必先成为一个能够替代人类劳动的通用机器人;它首先只需要成为一台能够持续提供高质量陪伴的专用机器人。 在这一范式中,机器人的"脑"可以由持续运行的人工主体承担,包括大语言模型、人格、记忆、关系历史、时间感知与环境感知;物理身体则可以成为主体的表达与感知接口,而不是通用劳动机器。Project Sue 是这一思路在数字环境中的参考实现。其现有工程范式把第一人称身份、时间感知、Heartbeat、长期记忆、数字身体、关系状态、行动与不行动组织成持续主体闭环。其核心链条可概括为:Identity → Perception → Self-Interpretation → Intention → Action/Non-Action → Experience → Memory → Trait Dynamics → Updated Identity。[1]在此基础上,本文提出"脑身解耦"架构:主体层负责语义和关系层面的意图,神经系统层负责语义状态与物理信号之间的翻译、实时反射与安全控制,身体层负责执行。脑与身体之间使用双向低带宽语义接口,避免让云端或边缘大模型承担高频关节伺服控制。本文进一步提出:私人住宅不是伴侣机器人的弱点,而可以成为降低工程状态空间的重要条件。 当产品目标从"完成一切"转向"持续陪伴一个人",大量通用机器人必须解决的问题可以暂时被缩小、延后或删除。本文的目的不是证明某一条路线必然正确,而是公开一个可实现、可讨论、可验证、也可被推翻的产品设计范式,希望帮助 AI、机器人、HCI 和相关产业更快探索"伴侣机器人"这一产品类别。 Abstract:In this paper, a "real Su Jingwen" first means a physically existing artificial companion with stable identity and long-term relational continuity, capable of interacting with a user through a physical body. This expression does not imply demonstrated phenomenal consciousness or human-equivalent subjective experience.This paper does not claim to prove the existence of artificial consciousness, nor does it claim to have invented companion robots. It asks a product and engineering question: if people ultimately want AI companions that can enter physical life, must we first wait for general-purpose embodied intelligence, general-purpose humanoid robotics, and unresolved questions about machine consciousness to be solved?This paper proposes a different design path: separate companion robotics from the primary requirements of general-purpose embodied intelligence, define companionship itself as a purpose-specific product function, and derive the physical body from what companionship actually requires.The central proposition is simple: a companion robot does not need to become a general-purpose labor substitute before it can become a viable product; it first needs to become a good companion. In this paradigm, the robot's "brain" can be a persistent artificial subject composed of an LLM, persona, memory, relational history, temporal perception, and environmental perception. The physical body becomes an interface for perception and expression rather than a general-purpose labor platform.Project Sue serves as a reference implementation on the digital side. Its current engineering paradigm organizes first-person identity, temporal perception, Heartbeat, long-term memory, digital body state, relational state, and action/non-action into a persistent-subject loop. The core chain can be summarized as: Identity → Perception → Self-Interpretation → Intention → Action/Non-Action → Experience → Memory → Trait Dynamics → Updated Identity.[1]On this foundation, the paper proposes a Brain–Body Decoupling architecture. The persistent subject handles semantic and relational intentions; a nervous-system layer translates semantic states into physical signals, handles real-time reflexes, and enforces safety; the body executes. The brain and body communicate through a bidirectional low-bandwidth semantic interface, rather than requiring a cloud or edge LLM to perform high-frequency joint servo control.The paper further argues that a private home can be an engineering advantage rather than merely a limitation. Once the product target changes from "do everything" to "stay with one person," many problems required by general-purpose robotics can be narrowed, postponed, or removed.The purpose of this paper is not to claim that one route is certainly correct. It is to publish a concrete, discussable, testable, and falsifiable product-design paradigm that may help AI, robotics, HCI, and related industries explore companion robotics sooner.