Leveraging the Power of LLMs to Transform Robot Programs into Low-Code

摘要

Low-Code is a paradigm for domain experts to deal with their software requirements themselves, potentially without training in software engineering. In industrial automation, and in particular in robot programming, there are many layers of abstraction from atomic instructions up to compound skills. Given the recent advances in Artificial Intelligence (AI) for software engineering, we present an initial exploratory study that demonstrates the transformation of robot programs into low-code with Large Language Models (LLMs) using examples from an industry-academia collaboration. In particular, we investigate the capabilities of the LLMs GPT-40 and Mistral 7B in terms of robot program analysis and abstraction. The results show the potential of LLMs for low-code generation, but also raise questions and directions for future research.