Energy-Aware Task Planning for Industrial Robots via Large Language Models

Abstract

Industrial robot task planning in manufacturing environments still relies heavily on manual programming and expert experience, which makes it difficult to respond quickly to flexible production requirements. Large language models (LLMs) provide a promising approach for translating natural language instructions into executable plans, but direct generation tends to fall short in domain grounding, execution rules, and practical constraints. This paper proposes an energy-aware task planning framework for industrial robots that integrates LLM-based task understanding with hierarchical knowledge support. The framework unifies graph-grounded retrieval, structured plan generation, and downstream robot program synthesis into a single workflow. It further introduces a four-layer knowledge organization mechanism covering action selection, strategy grounding, parameter setting, and constraint-aware planning. A prototype is implemented on an Epson industrial robot platform. Two representative case studies and a small-scale energy comparison on a handling task are conducted to evaluate the approach. Results show that the framework reliably interprets worker-oriented task requirements, produces structured and program-ready outputs, and supports downstream code generation. The energy experiment provides preliminary feasibility evidence that the proposed framework can support preference-aware planning and reduce total energy consumption in the tested handling scenario without compromising task success.