| Abstract [eng] |
Programming of industrial robots typically requires specialized knowledge of robot programming language, kinematics, and system configuration, which limits accessibility and slows deployment of robotic solutions. This paper proposes a methodology for intuitive robot programming using natural language instructions combined with large language model (LLM) based task planning. The proposed system employs a layered architecture consisting of a simulation environment, a Redis-based shared data layer, Python-based robot skills, a Behaviour Tree execution framework and an LLM-based planning module. Natural language instructions are converted into structured task plans that reference predefined robot skills and are executed through Behaviour Trees to ensure modular control and reliable failure handling. The system is evaluated in a simulated robot environment using CoppeliaSim, where 88 tasks with varying complexity are executed to assess planning reliability, execution success rate, and computational performance. Experimental results show that the proposed approach can successfully generate and execute task plans for most tested scenarios, particularly when instructions are clearly defined. The results demonstrate the potential of combining LLM-based planning with skill-based execution frameworks to support more intuitive and flexible robot programming. |
