Neuroergonomic signatures of improved human–robot collaboration in LLM-supported industrial workflows

Abstract

Abstract The Industry 5.0 is hindered by the communication bottleneck between humans and collaborative robots (cobots), because of the rigid programming requirements, which prevent dynamic and unexpected workflow changes. This study introduces and validates a Large Language Model (LLM)-enabled framework for natural language robot task planning to improve communication between humans and robots while simultaneously reducing the mental load of the worker. Utilizing a resource-efficient one-shot prompt engineering strategy, the system allows operators to adapt cobot trajectories via conversational input. We evaluated this dynamic workflow against a traditional static baseline in a simulated assembly environment, employing a comprehensive multimodal neuroergonomic assessment. Results demonstrate that the LLM framework achieved high technical efficacy, generating executable code within 1–2 prompts. Crucially, the combination of physiological metrics (heart rate, blink rate), behavioral data (task error rate), and subjective workload (NASA Task Load Index) revealed a significant reduction in operator strain and an increase in process reliability in the LLM-assisted condition. These findings provide empirical evidence that LLMs can lower the technical barriers to cobot programming while improving the physiological well-being of the workforce, thereby supporting a human-centric approach to flexible automation.