SemaPLC's AI agent writes 52% more reliable PLC code
New harness verifies AI-generated PLC code with live runtime tests, beating baselines by 20+ points
Researchers present SemaPLC, a project-grounded and verification-gated agent harness for PLC code generation. It declares a task complete only when logged external checks confirm the specification, the compilation, and the behavior on a live runtime. On 117 independent-POU tasks, it achieved the highest strict verified pass rate on all seven models, with a 72.6% mean. On a 65-task project-context track, it achieved the highest mean on integrated compilation, static behavior, and dynamic behavior. Dynamic behavior scores separated models sharply: baselines scored 22.4 to 31.4, while SemaPLC scored 52.2. The article concludes that execution, not static scoring, is the faithful test of whether generated control logic actually works. SemaPLC is open-sourced.
- SemaPLC is an open-source agent harness that generates and rigorously verifies PLC control logic using live runtime testing on actual PLC hardware.
- Achieved a 72.6% verified pass rate across 117 tasks — 20+ points higher than baselines on dynamic behavior tests (52.2 vs. 22.4–31.4).
- Verifies code via three layers: specification alignment, compilation success, and live runtime behavior matching — the latter being the most discriminative.
Why It Matters
Proves AI can safely generate industrial control logic when verified through real execution, reducing risk in automated manufacturing and infrastructure.