Title: Machine learning and biomechanical sensing toward real-time in-the-loop gait and joint health optimization

Date: Tuesday, April 29th 

Time: 11:00 am - 12:00 pm Eastern Time

Location: MRDC 4211

Zoom link: https://gatech.zoom.us/j/93148662032 (Meeting ID: 931 4866 2032)

Luis G. Rosa

Robotics PhD Candidate

Woodruff School of Mechanical Engineering

Georgia Institute of Technology

Committee:

Dr. Gregory Sawicki (Advisor) - School of Mechanical Engineering & School of Biological Sciences, Georgia Institute of Technology

Dr. Omer Inan (Advisor) - School of Electrical and Computer Engineering, Georgia Institute of Technology

Dr. Aaron Young - School of Mechanical Engineering, Georgia Institute of Technology
Dr. Maegan Tucker - School of Mechanical Engineering & School of Electrical and Computer Engineering, Georgia Institute of Technology

Dr. Frank Hammond III - School of Mechanical Engineering, Georgia Institute of Technology

Abstract:

Despite advances in wearable sensing and assistive devices, current systems often rely on indirect or delayed signals that limit their ability to capture subcutaneous physiological dynamics in real-time. This challenge hinders progress across domains ranging from exoskeleton control to clinical monitoring in populations with movement or inflammatory disorders. This work aimed to expand corresponding biomechanical sensing capabilities by developing a novel sensing framework capable of extracting under-the-skin biomechanical signals from muscle and tendon structures using ultrasound and active acoustics. To achieve this, we (Aim 1) developed a machine learning pipeline for real-time estimation of muscle fascicle lengths from B-mode ultrasound images to enable “muscle-in-the-loop” feedback systems; (Aim 2) introduced and benchmarked an active acoustics sensor capable of measuring Achilles tendon loading in real-time with low latency across a wide range of locomotion tasks; and (Aim 3) applied our acoustics sensing approach in a pediatric arthritis cohort to quantify how inflammation-related physiological alterations affect machine learning task classification performance, highlighting its potential as a non-invasive biomarker for disease presence and severity. Collectively, these studies establish a new direction for non-invasive, task-relevant muscle-tendon sensing that could inform next-generation systems for rehabilitation, augmentation, and clinical