Machine Learning Methods to Individualize Powered Orthotic Intervention for Improved Functional Recovery After Lower Extremity Trauma
Abstract
The long-term objective of this project is to improve the outcomes of robot-assisted exercise interventions for limb salvage patients who suffered HELETs and facilitate return to work/duty. The project will fill current gaps in powered AFO technology by establishing novel ML methods to enable patient tailored AFO designs and self-adaptive active AFO assistance. Within the current reporting period, the project team has developed an efficient AFO design workflow, which leverages low-cost laser scan technology, open-source CAD software, and AM processes to generate orthotic designs that conform to the leg morphology while requiring minimal labor. After securing IRB approval from USAMRMC, the workflow was validated by 3Dprinting subject-tailored AFOs for 10 able-bodied individuals. Bench tests are underway to characterize mechanical properties of the new AFOs, following which the team will carry out comfort tests with able-bodied individuals and HELET patients. Concurrently, the team has developed a removable, lightweight, high-performance cable-driven actuator and implemented a closed-loop controller that demonstrated excellent torque bandwidth under a wide range of loading conditions. Using ML stochastic models, the team also developed a subject-agnostic estimator of the biological ankle moment, which was tested at different walking speeds during treadmill tests with 10 able-bodied individuals. In near future, the ankle-moment estimator and the torque controller will be integrated into a reinforcement-learning (RL) assistive controller for the AFO, to be tested in able-bodied individuals and HELET patients.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 2023
- Accession Number
- AD1216914
Entities
People
- Damiano Zanotto
- Karen J. Nolan
- Kishore Pochiraju
Organizations
- Kessler Foundation
- Stevens Institute of Technology