ABSTRACT Chronic Low Back Pain (CLBP) is a complex multi-factorial condition, as well as the most prevalent painful musculoskeletal disorder worldwide. Identifying the optimal treatment for CLBP on a patient-specific basis is an important and unresolved challenge in medicine. Tailoring interventions according to patient movement characteristics may improve clinical outcomes. Patients with CLBP are heterogenous in terms of their symptoms, clinical exam findings, and conventional medical imaging results. For most patients, the optimal treatment plan is unknown, therefore it is challenging for the clinician to prescribe an appropriate and cost- effective course of treatment. One important clinical characteristic that can be used for classification is severity of physical impairment (problems in lumbar spine structure and function) and resulting activity limitation (difficulty executing activities). A common approach to assess the impact of physical impairment is using patient-reported outcomes (PROs), wherein patients rate their perceived ability to perform various activities in their usual environment. PROs are subjective and discrepancies have been observed between how patients score PROs and how they perform activities when observed in the clinic. It is advantageous to complement PROs with objective performance-based measures of physical function. Therefore, the overall hypothesis of the Biomechanical Core of the parent grant is that including patient-specific spine biomechanics in predictive models improves our ability to characterize CLBP patients. To that end, the purpose of this administrative supplement is to expand upon Specific Aim 2 of the Biomechanical Core, which is to characterize lumbopelvic kinematics during functional tasks and daily activities using wearable (inertial) motion sensors. Specifically, this work will aim to develop deep (machine) learning algorithms that can correctly identify and characterize motions of the lumbar spine during both clinical and field assessments. During the clinical assessments, participants will be asked to perform functional tasks while wearing inertial measurement units (IMUs). Collected data will be used to develop and train machine learning algorithms to identify tasks of interest such as activities of daily living and aberrant/painful motions. The deep learning algorithms developed will be used to label lumbar motion data collected continuously during field assessment in patients' homes over a 7-day testing period. The supplemental data will be compared with the standard data analyses approaches proposed for the overall study and included with the LB3P phenotyping. Moreover, the deep learning algorithms will serve as the foundation for the development of ecological momentary interventions that are responsive to patient's real-world functional impairments related to CLBP.