具身智能动作决策的最小证据边界 / Minimal Evidence Boundaries for Action Decisions in Embodied Agents
Year: 2026
Type: preprint
Abstract: 本稿为中文方法与理论研究草稿,提出以当前动作是否可被证据确认,作为进一步获取观测的停止条件。稿件区分静态离线最短证书与在线自适应查询代价,并给出基本推导、动态失效反例与后续比较实验方案。研究仍处于探索阶段,新颖性与实际工程收益尚未建立;本文不是经同行评审的期刊论文。 Minimal Evidence Boundaries for Action Decisions in Embodied Agents. This Chinese-language methods and theory draft studies action-specific evidence acquisition, distinguishing offline certificates from online adaptive query costs. It includes elementary propositions, a dynamic invalidation counterexample, and a falsifiable evaluation plan. Novelty and practical benefits remain to be established. This is an exploratory preprint, not a peer-reviewed journal article. Validation is limited to a four-bit static logical model: 16 worlds, 81 partial-evidence states and 256 compatible world/evidence pairs. The included standard-library Python program and actual JSON output reproduce these checks. There are no physical robot experiments, learned-model benchmarks or claims of real-world safety. Query-count comparisons apply only to the stated toy model and uniform distribution. Files: six-page Chinese manuscript PDF, editable DOCX, Python checker, actual enumeration results and reproduction instructions. Manuscript version completed on 2026-11-13; repository publication date is 2026-11-14. The manuscript's statement that it had not yet been submitted describes its status when written. AI assistance was used for ideation, literature discovery, drafting, formalization, coding and typesetting. Related work and limitations are explicitly discussed. The listed author and affiliation follow author-provided information: YANGZHOU, 梗必爆具身交互AI研究院 (Genbibao Embodied AI Research Institute), ORCID 0009-0005-4736-2097. This record does not assert journal acceptance or indexing by CNKI or Scopus.
Author: YANGZHOU
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225983447
