Development and Evaluation of a Machine Learning Model for Cobot Action Prediction: Optimizing Algorithms Through Cognitive Workload Metrics

Collaborative robots (Cobots) have replaced traditional industrial robots on the factory floors, enhancing automation and efficiency. From the perspective of Industry 5.0, centering around H.F. (Human Factors), concerns about cognitive ergonomics persist as Cobots' unpredictable actions can increase cognitive workload and reduce productivity. Identifying and prioritising the operators' mental and physical well-being is necessary for safer and more efficient Human Robot Collaboration (HRC). This paper proposes a novel machine-learning model for Cobot action prediction and evaluates which algorithm is best for predicting Cobot actions based on operators' cognitive states. 75 participants with different levels of technophobia, technical expertise, and age collaborated with a Cobot to solve a complex maze puzzle. Firstly, each participant's Cognitive Workload (CWL) was measured using a combination of subjective and objective metrics (NASA-Task Load index, Heart Rate Variability, and Error Rate), followed by the development of a machine learning model that takes human factors as input parameters and predicts Cobot's actions accordingly. This study evaluates various machine learning models for Cobot action prediction and identifies that the Random Forest Classifier outperforms other models with an accuracy of 90%. The model can be implemented in various industrial settings to improve HRC safety and operators' cognitive ergonomics.