SemaPLC: A Project-Grounded, Verification-Gated Agent Harness for PLC Code Generation
Source
arxiv.org
Author
Yanlun Tu, Huacan Wang, Ziyue Zhou, Jie Zhou, Ningyan Zhu, Ge Chen, Wangyi Chen, Tengfei Zhou, Yifan Zhou, Dasheng Yang, Xiaofeng Mou, Hui Zhang, Yi Xu
Date
Context
SemaPLC makes externally checked behavior the stopping condition for an AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → writing industrial-controller code. Its checks cover the specification, compilation and execution, so the model cannot declare success solely from inspecting its own output. In the authors' reported evaluation, generated programs and reference programs ran on a live programmable logic controller runtime, and their execution traces were compared. Across 65 tasks requiring integration into existing projects, runtime results separated the methods much more sharply than static checks did. The finding supports testing what generated control logic actually does within a project, beyond whether the code compiles.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.