ECHO: A Natural Language-Enabled Cognitive Framework for Human-Centric Energy Optimization of Industrial Robots

Abstract

Optimizing the energy efficiency of industrial robots is key to sustainable manufacturing. However, many fully autonomous systems in the literature tend to overlook the dynamic realities of the production floor and exclude production operators from the decision-making loop by requiring robot programming knowledge. This study presents ECHO (Energy-efficient Cognitive Human-centric Optimization), a Human-in-the-Loop (HITL) energy optimization framework with a Large Language Model (LLM)-based natural language interaction layer, in line with the human-centric vision of Industry 5.0. The proposed ECHO framework comprises three layers: algorithm competition, natural language-based parameter revision, and real-time energy validation. ECHO was validated across three scenarios on a 6-degree-of-freedom industrial manipulator. Experimental results across ten independent runs per algorithm showed that the leading algorithms achieved average energy reductions of 29.7–30.2% and cycle-time reductions of 28.0–28.5% across the three scenarios, compared with expert-defined baseline operation. Bayesian Optimization converged fastest and most consistently under the limited operational budget, while Particle Swarm Optimization and Reinforcement Learning reached comparable performance as the optimization progressed. A user study with 16 shop-floor personnel without prior robotic programming experience yielded a System Usability Scale (SUS) score of $83.6 \pm 7.0$ and a NASA Task Load Index (NASA-TLX) score of $21.1 \pm 7.5$ . These results indicate that ECHO connects algorithmic optimization with shop-floor expertise, enabling complex energy tuning without robot programming knowledge.