LOADX Dataset: workLOad Assessment in Dual-task eXecution through multimodal monitoring

LOADX (workLOad Assessment in Dual-task eXecution) is a multimodal dataset developed to support research on physical and mental workload estimation during single- and dual-task activities. The dataset contains synchronized physiological, kinematic, task-related, and subjective data collected from 20 participants performing a structured experimental protocol designed to elicit different combinations of physical and cognitive demand. The protocol includes 16 experimental conditions: three physical tasks, three cognitive tasks, nine dual-task combinations, and one resting baseline. The physical tasks consist of repeated shoulder flexion–extension, marching in place, and squats. The cognitive tasks include a Stroop test, a 2-back working-memory task, and backward counting by seven. Each active condition lasted 90 seconds and was followed by a resting interval. Physiological data were acquired using wearable sensors and include raw electrocardiographic signals, raw respiratory waveforms, heart rate, inter-beat interval, respiratory rate, galvanic skin response, and triaxial acceleration. Full-body kinematic data were collected using a contactless vision-based monitoring system based on MediaPipe Pose, providing the three-dimensional coordinates of 33 anatomical landmarks. All acquisition streams share a common temporal reference, enabling the alignment of physiological signals, body movements, task execution, and workload annotations. After each experimental condition, participants rated their perceived physical and mental demand using selected dimensions of the NASA Task Load Index. The dataset therefore includes task-level Physical Demand and Mental Demand scores, which can be used as subjective workload annotations for supervised learning, correlation analysis, clustering, and validation of multimodal workload-estimation methods. The dataset is organized into one folder for each participant, identified through an anonymized numerical code such as S01, S02, and so forth. Each participant folder contains: Sxx.csv: synchronized physiological, kinematic, task-related, and subjective data; Sxx_ecg.csv: raw ECG waveform; Sxx_breath.csv: raw respiratory waveform; Sxx_nasa.xlsx: task-level NASA-TLX Physical Demand and Mental Demand ratings. LOADX can support research on multimodal workload assessment, physiological and movement-signal processing, sensor fusion, feature extraction, task recognition, modality-ablation analysis, subject-independent validation, and the development of adaptive human–machine interaction systems. The inclusion of physical-only, cognitive-only, and dual-task conditions enables the investigation of how physical and mental demands interact and how they are reflected in complementary sensing modalities. LOADX should be considered a structured multimodal research dataset rather than a population-level normative dataset. Inter-individual variability was intentionally retained to support the investigation of personalized and subject-independent workload-estimation approaches.