发表机构
Hezarfen LLC; Grand Canyon University（赫扎芬有限责任公司; 大峡谷大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
提出CT-SAFR多层验证框架，通过思维链推理检测与过滤，实现94.2%幻觉检测率和87%不安全输出减少，保障自主机器人决策的安全性与可解释性。
AI 中文摘要
思维链（Chain-of-Thought，CoT）提示使大型语言模型（LLM）能够进行显式、逐步的推理，为复杂的自主机器人创造了机遇。然而，近期研究表明，推理模型仅25-39%的时间会将其实际决策过程言语化，在复杂任务上忠实度下降44%。本文提出CT-SAFR（面向机器人的思维链安全性与忠实度），一种多层验证框架，实现了94.2%的幻觉检测率（n=500，95%置信区间：91.8-95.9%），延迟低于500毫秒。通过仓库机器人案例研究，本工作展示了不安全推理输出减少87%（p<0.001），并为具备推理能力的自主机器人的负责任部署提供了建议。
英文摘要
Chain-of-Thought (CoT) prompting enables LLMs to perform explicit, step-by-step reasoning, creating opportunities for sophisticated autonomous robots. However, recent research reveals that reasoning models verbalize their actual decision processes only 25-39% of the time, with faithfulness degrading 44% on complex tasks. This paper presents CT-SAFR (Chain-of-Thought Safety and Faithfulness for Robotics), a multi-layered verification framework achieving 94.2% hallucination detection (n = 500, 95% CI: 91.8-95.9%) with sub-500ms latency. Through a warehouse robot case study, this work demonstrates 87% reduction in unsafe reasoning outputs (p < 0.001) and provides recommendations for responsible deployment of reasoning-capable autonomous robots.
Comments7 pages, 4 figures. Accepted version. Published in 2026 IEEE Conference on Artificial Intelligence (CAI), pp. 598-603
Journal ref2026 IEEE Conference on Artificial Intelligence (CAI), pp. 598-603, May 2026
DOI：10.1109/CAI68641.2026.11536646
