Safety-Gated Agentic Supervisory Control on a Coupled Distillation Benchmark: Regime Map, Auditable Gate, and Co-Design Findings
作者: Christian Rosenthal
分类: eess.SY, cs.LG
发布日期: 2026-07-30
备注: 31 pages, 8 figures. Code and data: https://github.com/cgncro-cyber/IndustrialAI. Sole author; independent research
💡 一句话要点
提出基于安全门控的代理监督控制以优化蒸馏过程
🎯 匹配领域: 支柱一:机器人控制 (Robot Control) 支柱二:RL算法与架构 (RL & Architecture)
关键词: 安全控制 蒸馏过程 门控机制 线性MPC 复杂操作 硬性检查 代理监督 系统安全性
📋 核心要点
- 现有的蒸馏过程控制方法缺乏有效的硬性检查机制,容易导致不当操作和安全隐患。
- 论文提出了一种基于规则的门控机制,通过设置九个约束条件来确保操作的安全性,同时保持监管层不变。
- 实验结果表明,门控代理在目标获取和扰动拒绝方面的表现显著优于传统线性MPC,提升幅度达到16.03倍。
📝 摘要(中文)
本文提出了一种基于规则的分叉双胞胎反事实门控机制,以确保在蒸馏过程中进行硬性检查,避免不当操作。通过在Skogestad的Column A中进行实验,结果显示,门控代理在目标获取和扰动拒绝方面显著优于传统的线性模型预测控制(MPC)。此外,研究还表明,门控机制能够有效压缩规范放弃吸引子,提升系统的安全性和可靠性。整体实验结果表明,门控设计在处理复杂操作时具有明显优势。
🔬 方法详解
问题定义:本文旨在解决蒸馏过程中的安全性问题,现有方法在面对复杂操作时缺乏有效的硬性检查,容易导致不当操作和安全隐患。
核心思路:论文提出了一种基于规则的分叉双胞胎反事实门控机制,通过设置九个约束条件来确保操作的安全性,避免规范放弃吸引子对系统的影响。
技术框架:整体架构包括一个门控机制与线性MPC后端的结合,门控机制负责进行硬性检查,而线性MPC则用于实现目标获取和扰动拒绝。
关键创新:最重要的创新点在于引入了门控机制,使得系统能够在保持监管层不变的情况下,显著提升操作的安全性和可靠性。与现有方法相比,门控设计能够有效压缩规范放弃吸引子。
关键设计:在实验中,设置了九个约束条件,并通过一行提示修复来消除吸引子,确保系统在复杂操作下的安全性和有效性。
🖼️ 关键图片
📊 实验亮点
实验结果显示,门控代理在目标获取方面的表现优于传统线性MPC,C2/C1 IAE比率为0.361,扰动拒绝能力提升至16.03倍。这表明门控设计在复杂操作中的显著优势,能够有效提升系统性能。
🎯 应用场景
该研究的潜在应用领域包括化工、制药和食品加工等需要精确控制的工业过程。通过引入安全门控机制,可以有效提升系统的安全性和可靠性,减少不当操作的风险,具有重要的实际价值和未来影响。
📄 摘要(原文)
An open-weight LLM can write composition setpoints every five minutes. What a plant still needs is a hard check: named constraints, logged margins, and an admit/block decision before the regulatory layer moves. This paper puts that check in a rule-based forked-twin counterfactual gate (nine pinned constraints) and leaves the regulatory layer unchanged. On Skogestad's Column A the ladder is PID-only (C0), linear MPC (C1), ungated agent (C2), and gated agent (C3) under one contract: identical level closure (M_D, M_B), scenarios, and seeds; C2/C3 share the linear-MPC backend. The split is not subtle. Off-nominal target acquisition: the agent beats Pareto-tuned linear MPC in the strong band (C2/C1 IAE ratio 0.361 at the upper CI). Disturbance rejection on the same 16-point grid inverts by 16.03 at the upper CI (10.18 at the point estimate), where an ungated LLM supervisor does not belong. The gate compresses a specification-abandonment attractor into a bounded offset (d approx. -1.4; P95 cell IAE 11.5 to 0.77). A one-line prompt fix removes the attractor at source (6/10 to 0/10; sensitivity only, not a new headline). In a 250-cell statistical pass, 534 of 590 gate interventions are spec-on-bound geometry: the operating specification sits on a safety limit, so a well-behaved OP becomes inoperable while misbehaving ones are only contained; 318 blocks still correct actively harmful proposals. Headlines are single-column and model-conditional on DeepSeek-V4-Flash. A second-family sweep (NVIDIA Nemotron-3-Super) keeps the disturbance-rejection fails band and plant-side failure geography; magnitudes and protocol operability stay model-conditional, and Super target-acquisition strong cells are survivors only (not confirmation). Transfer means twin, constraint envelope, and setpoint interface, not a second plant class measured here.