A Use Case of Natural Language Programming for Industrial Robots

Recent advances in Large Language Models (LLMs) have enabled their use in natural language tasks across various domains, including robotics. Recent studies have extensively explored the integration of LLMs with social robots, highlighting how their ability to generate human-like language can lead to more natural, coherent, and effective interactions with people. In industrial settings, combining LLMs and other AI technologies with collaborative or industrial robots can support the development of more intelligent robotic behaviors, ultimately improving task execution efficiency. Previous research has demonstrated that it is possible to integrate LLMs and vision-based LLMs to perform robotic tasks described using natural language. Moreover, LLMs have demonstrated the ability to convert natural language into low-level actions that enable the execution of industrial tasks. LLMs have limitations, such as biased outputs, hallucinations, lack of context, and no persistent memory, making adaptation to specific environments challenging. To address this, the authors previously developed a cognitive architecture using LLMs and common sense knowledge to process semi-structured data via modular components (Supervisors), utilizing prompt engineering. Although this architecture was originally designed for general interaction scenarios, in this work, we aim to explore its integration within an industrial context. Unlike social environments, which require the robot to be capable of handling social behaviors and intents, here the robot is required to understand and execute complex tasks. To achieve its goals, the robot must be capable of understanding the human expert’s requirements and feedback expressed in natural language. It must also be capable of reason and understand how to comply with specified constraints and avoid diverging from the human-imposed goal.