Large Language Models and Adversarial Reinforcement Learning to Automate PLCs Programming: A Preliminary Investigation

The manual programming of Programmable Logic Controllers (PLCs) is a time-consuming and error-prone task, particularly for complex systems. Researchers have proposed various techniques, such as the modular, decentralized, hierarchical, and distributed approaches, which all fall under the Supervisory Control Theory (SCT) whose main goal is to generate a supervisor that ensures that the controller's behavior satisfies the specifications. While these techniques have demonstrated efficiency in systems of low complexity, high-complexity systems and probabilistic ones remain a major challenge. In this paper, we propose a novel pipeline that relies solely on Artificial Intelligence (AI)-techniques, namely, Reinforcement Learning (RL) and Natural Language Processing (NLP) techniques, including Named Entity Recognition (NER) and Large Language Models (LLMs). As a preliminary result, we demonstrate the potential of the proposed pipeline using the recent ChatGPT model by making a sample dataset in this context and generating Structured Text code from specifications presented in a natural language format. Finally, we outline future directions toward a fully automated AI-based PLC programming tool.