Title: Enabling Scalable, Robust, and Versatile Exoskeleton Control Using Deep Learning and Transfer Learning Methods 

Date: Tuesday, November 15th

Time: 9AM EST

Location: GTMI Auditorium (Room 101)

Keaton Scherpereel

Robotics PhD Student

School of Mechanical Engineering

Georgia Institute of Technology

Committee:

Dr. Aaron Young (Advisor) – School of Mechanical Engineering, Georgia Institute of Technology

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

Dr. Matthew Gombolay – School of Interactive Computing, Georgia Institute of Technology

Dr. Thomas Ploetz – School of Interactive Computing, Georgia Institute of Technology

Dr. Gregory Sawicki – School of Mechanical Engineering, Georgia Institute of Technology

Abstract:

Current state-of-the-art exoskeleton control is tailored to aid users in only one specific task or a finite number of tasks, but this task specificity hampers real-world application due to the variable and sporadic nature of human movement. A novel and emerging solution to this inherent drawback is to use deep learning models with inputs from wearable sensors to directly estimate user’s internal biological joint moment. This provides a continuous, task-agnostic signal upon which to build an assistance profile. To accelerate the advancement of this new approach, unique sensing modalities must be explored to better capture internal physiological states (Aim 1), additional data must be leveraged to improve user and task generalizability (Aim 2), and a path must be established for scaling this control architecture to novel devices (Aim 3). These advancements will be a useful step in propelling exoskeleton technologies beyond laboratory testing to real-world applications and contribute to scientific knowledge in exoskeleton control and applied machine learning.