Systematic Translation from Natural Language Robot Task Descriptions to STL

We present a systematic pipeline that converts free-form natural language descriptions of robot tasks into Signal Temporal Logic (STL) specifications. Leveraging large language models for semantic parsing, plus a formal post-processing stage that enforces syntactic and semantic STL constraints, our approach yields machine-verifiable goals from unstructured instructions. Experiments on a diverse benchmark of household-robot scenarios show a significant improvement in translation accuracy and downstream task execution success over existing rule-based methods.

In AISoLA 2024 — Artificial Intelligence Applications & Innovations