A Kinematic and Electromyographic Dataset for Ergonomic Risk Assessment in Manual Material Handling

Automated ergonomic exposure and risk assessments, leveraging measures from wearable sensors and markerless motion capture systems, is being increasingly adopted in manual material handling contexts as an alternative to self-reports and direct observation. Machine learning and artificial intelligence (ML/AI) algorithms trained on these measures enable continuous estimation of musculoskeletal demands and injury risk; however, model accuracy and generalizability remain constrained when training datasets lack diversity in worker characteristics and task conditions. In particular, insufficient representation of demographic factors and load-handling scenarios can lead to biased or non-generalizable models. Existing open-access, occupational movement datasets rarely capture such variability, often relying on small, homogeneous samples and single load types. To address these limitations, we present a cloud-based, public database comprising data from 40 participants (20 male, 20 female), including younger adults [mean (SD) age = 23 (3.1) years] and older adults [55.5 (5.3) years], who performed simulated occupational lifting/lowering and pushing/pulling tasks. Data were collected from three sensing modalities—inertial measurement units (IMUs), markerless motion capture (Kinect), and an electromyography (EMG) armband—using multiple load configurations, load masses (2.3-11.3 kg), lifting heights, and handling depths. This database supports the development, benchmarking, and cross-modal evaluation of ML/AI algorithms for ergonomic exposure and risk assessments in occupational settings.