A multimodal EMG and IMU dataset for assessing the quality of exercises designed for spatially constrained environments.

Regular physical training is essential for maintaining physical fitness in environments where exercise must be performed in limited space and without direct professional supervision. In such settings, degraded exercise execution may reduce training effectiveness and increase injury risk. To address this challenge, this study aimed to develop a methodology for collecting and processing multimodal data to support autonomous, AI-driven feedback systems. We collected data from 20 healthy adults performing a structured protocol of specialized, whole-body exercises designed for spatially constrained conditions. The dataset integrates surface electromyography (sEMG) signals from four key muscles, body kinematics, and wrist-based heart rate. Raw sensor data were processed, filtered, and segmented. The final dataset includes synchronized EMG envelopes, IMU-derived kinematic features (quaternions), heart rate data, and expert annotations provided by physiotherapists who evaluated movement quality against predefined biomechanical criteria. This comprehensive dataset is designed to facilitate the development and validation of machine learning models for automated exercise quality assessment in settings where space is limited and direct supervision is unavailable.