Wearable sensing technologies offer promising solutions for unobtrusive, continuous monitoring of cognitive fatigue (CF) during human–robot interaction (HRI). This paper presents a novel dataset collected from wearable electrocardiography (ECG) and electrodermal activity (EDA) sensors during collaborative HRI tasks and evaluates their effectiveness for low, medium, and high CF classification. Physiological features were extracted from both modalities and used to train supervised machine learning models. The best-performing model, a Gradient Boosting Machine (GBM), achieved a mean accuracy of 95.86% and an F1 score of 95.75%, demonstrating strong sensitivity across all fatigue levels. ECG-based features were particularly discriminative, while combining ECG with EDA offered slight gains in robustness, especially for medium fatigue detection. These results highlight the feasibility of using lightweight wearable sensors for accurate CF assessment and support their integration into fatigue-aware, adaptive support systems for assistive robotics. This work contributes to the growing field of wearable cognitive state monitoring.