paperbot · PL 论文追踪

全部论文 待读 收藏

P2 Dynamic update for synthesized GR(1) controllers

合成 GR(1) 控制器的动态更新
ICSE Proceedings of the 44th International Conference on Software Engineering2022匹配度 3/10 · 论文属于反应式综合/形式化验证领域的 GR(1) 控制器动态更新,虽具有形式化定义、工具实现和基准评估,但与用户核心关注的 LLM/Agent 与静态分析、形式化验证的交叉,以及浮点/贝叶斯程序分析均无直接关联。引用 6 inbox
Gal Amram, Shahar Maoz, Itai Segall, Matan Yossef
crossref: 2026-09-12dblp: 2026-09-12

AI 速览

一句话:提出并求解 GR(1) 合成控制器的动态更新问题,用桥接控制器实现运行中安全切换至新规约。

反应式合成可从时序逻辑规约自动构造正确性内建的反应系统,GR(1) 是其中可高效求解的 LTL 片段。本文研究 GR(1) 的动态更新问题:让已运行中的合成控制器在不停止的前提下,安全地转为满足修改后的最新规约。作者形式化定义了该问题,并提出一种可靠且完备的解法,其核心是计算一个桥接控制器(bridge-controller),以协调旧、新控制器行为并完成切换。该工作已在 Spectra 合成与执行环境中实现,并在基准规约上评估;结果表明动态更新具有较好的效率与有效性,推进了反应式合成的应用边界。

问题
对已运行中的 GR(1) 合成控制器,如何在不停机的情况下安全地更新其行为,使其符合修改后的规约?
方法
形式化定义 GR(1) 动态更新问题,提出基于 bridge-controller 计算的可靠且完备解法,并在 Spectra 环境中实现。
评估
在基准规约上评估实现,验证动态更新方案的效率与有效性。
原文摘要(Abstract)

Reactive synthesis is an automated procedure to obtain a correct-by-construction reactive system from its temporal logic specification. GR(1) is an expressive fragment of LTL that enables efficient synthesis and has been recently used in different contexts and application domains. In this paper we investigate the dynamic-update problem for GR(1): updating the behavior of an already running synthesized controller such that it would safely and dynamically, without stopping, start conforming to a modified, up-to-date specification. We formally define the dynamic-update problem and present a sound and complete solution that is based on the computation of a bridge-controller. We implemented the work in the Spectra synthesis and execution environment and evaluated it over benchmark specifications. The evaluation shows the efficiency and effectiveness of using dynamic updates. The work advances the state-of-the-art in reactive synthesis and opens the way to its use in application domains where dynamic updates are a necessary requirement.

链接与引用

DOI 原文 · PDF(开放获取) · DBLP

BibTeX
@inproceedings{AmramMSY22,
  title = {Dynamic update for synthesized GR(1) controllers},
  author = {Gal Amram and Shahar Maoz and Itai Segall and Matan Yossef},
  booktitle = {ICSE},
  year = {2022},
  doi = {10.1145/3510003.3510054}
}