A Stem-Cell Developmental General-Purpose Kernel for Cross-Embodiment Adaptation: An Endogenous Developmental Architecture
This record contains the English and Chinese versions of the same conceptual preprint. No implementation or empirical results are reported. Cross-hardware transfer in embodied AI commonly relies on multi-embodiment pretraining, action-space alignment, downstream adaptation, or explicit model identification. This article proposes a complementary route: a Stem-Cell Developmental Agent (SCDA) kernel that separates a persistently existing cognitive organization from interchangeable physical embodiments. The same kernel is regulated across a continuous, non-episodic individual history by four endogenous norms: homeostatic maintenance, exploratory differentiation, autobiographical consolidation, and self–world boundary learning. It first develops the capacity to acquire new body models across diverse simulated morphologies and subsequently connects to previously unseen embodiments through a standardized hardware abstraction interface. Cross-embodiment adaptation is specified as four observable stages: interface validation, constrained active identification, causal boundary incorporation, and capability expansion. Core norms and individual history persist across attachment events, whereas embodiment-specific models can be created, revised, suspended, or revoked. We define the kernel–embodiment state model, specify interface fields and safety responsibilities, and propose five control conditions, five falsifiable hypotheses, and metrics covering adaptation efficiency, identity continuity, fault degradation, and safety violations. The proposed contribution is not the first method for cross-robot transfer, nor a claim that one kernel can unconditionally control arbitrary hardware. It is a narrower test of whether persistent subject-level organization improves cross-embodiment adaptation and recoverability. Self-boundaries and autobiographical narratives denote measurable computational functions, not phenomenal consciousness.
Authors
- Yuan Ye
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-02
- DOI
- https://doi.org/10.5281/zenodo.23874261
- Primary Topic
- Psychiatry, Mental Health, Neuroscience
- Type
- article
- Field-Weighted Citation Impact
- 0.00