Collaborative LLM agents for flexible software development of intelligent industrial robot control systems

Abstract

Software plays a crucial role in robot control systems, and its efficient, flexible development is essential for production. Such software must be customized to specific production processes, requiring developers with specialized expertise in these areas—the high development threshold results in reduced efficiency and increased costs. Recently, significant progress has been made in automated problem-solving through societies of agents based on large language models (LLMs). To automate software development for industrial robot control systems, this paper introduces an Industrial Robot Control Software Auto-Development (IRCSAD) framework with multi-agent collaboration, and a Low-code Industrial Software Platform (LISP) for validating IRCSAD-developed software. IRCSAD automates software development and iterative optimization of prompts through the collaboration of multiple LLM-based agents. This work also proposes a software testing methodology for robot control systems based on reliability assessment. An experimental study that develops software for the robot control system for assembly, sorting, and inspection tasks is implemented. The results show that collaboration with LISP enhances IRCSAD's ability to solve complex problems in the development process, saves development costs, and improves development efficiency.