A Multimodal Dataset for Stress Detection in Collaborative Physical Human-Robot Interaction

This work presents the Collaborative Stress (CoStress) dataset, a multimodal dataset designed to investigate stress dynamics during collaborative physical Human-Robot Interaction (pHRI). The dataset includes synchronized recordings from 20 healthy participants performing three-dimensional point-to-point reaching tasks with an end-effector robotic device. It provides robot-side measurements such as end-effector kinematics, interaction forces, and assistance levels, together with multimodal physiological signals including heart rate, respiration activity, and galvanic skin response, as well as subjective assessments of stress and workload collected through standardized questionnaires (SUDS and NASA-TLX).Data were collected under controlled experimental conditions in which the level of robot guidance was systematically modulated by varying the stiffness value of the interaction controller, thus inducing different physical and cognitive demands during task execution. By combining robot-side measurements acquired during collaborative pHRI, including interaction forces, kinematics, and assistance levels, with synchronized physiological and subjective data, CoStress enables the study of how stress and workload influence motor performance and physical interaction under different assistance conditions. The dataset addresses the current lack of publicly available resources specifically designed to investigate stress-aware adaptation in collaborative pHRI scenarios.Beyond serving as a benchmark for stress detection, CoStress supports research in multimodal signal processing, feature extraction, and machine-learning-based modeling of user state during pHRI. More broadly, the dataset is intended to facilitate the development of adaptive and human-aware robotic systems capable of personalizing interaction strategies according to the user’s current physical and cognitive condition.