Researchers have developed intelligent robotic systems that adapt to individual learning patterns, potentially transforming the outcomes of physical therapy for patients with mobility disorders
The path to recovery for stroke survivors is often overwhelming for individuals who take the first interim measures or navigate daily challenges from balance problems. As people walk, their brains and bodies need to work together to coordinate complex processes that involve multiple muscles and joints to maintain balance and continue moving forward.
A new research project led by FAMU-FSU University of Engineering Assistant Professor Taylor Higgins We investigate the process of athletic ability by studying how people learn to control machines where balance and movement are constant and important.
This study has three main goals: To study how humans learn to balance and propel on a unicycle. Comparing human learning with machine learning. These discoveries are then integrated into robotically assisting devices that accelerate the speed at which humans learn new movements.
Researchers want to use unicycle robots and accompanying algorithms to study how robot-based platforms can help humans acquire skills. This work will inform future research and physical therapy tools that they can relearse to relear patients.
“Many supporting robot research focuses on supporting people who have movements that they already know how to play,” Higgins said. “Our research focuses on how people acquire new athletic abilities. If robots can help healthy people learn tasks faster, they can help people undergoing rehabilitation regain lost skills faster.
The $799,000 project is supported by the National Science Foundation Mind, Machine, Motor Nexus Programor m3x.
The M3X program supporting this project funds basic research that enables humans to interact with intelligent engineering systems. This research requires advanced technology to respond to this data by collecting data from human users and adapting its behavior on the fly. The goal is to increase safety, productivity and well-being for people adopting robotic assistance in complex and changing circumstances.
How it works
Creating mathematical models for walking is complicated. This movement involves repeatedly creating and destroying contact with the ground in what engineers call hybrid dynamic systems. In contrast, the unicycle remains in contact with the ground. This represents a continuous system that is easy to model mathematically.
“It's really hard to get a robot to walk on two legs,” Higgins said. “This problem becomes so difficult when you need to guide the learning process simultaneously. It uses simpler mathematical models that allow you to focus more on how the robots are involved with humans.”
The first step of the project is to study how subjects learn about unassisted unicycles. As new riders practice, the researchers use motion capture suits to collect data from co-locations and mathematically model movement through space during the learning process.
Researchers then use reinforcement learning to train simulated “robot coaches” to guide simulated human learning to a unicycle. Coaches are rewarded by accelerating the progress of learners' skills. Although each agent is simulated, this part of the project explores how reinforcement learning promotes two-way relationships between robotic coaches and human learners. The coach guides the learners and informs the coach of their progress.
Finally, Higgins' team will build a unicycle that can provide robotic assistance and teach beginner riders using learning strategies developed in Step 2. They study whether robotic coaches accelerate the learning process.
The coaches will assist: initially assist the learners, then reduce support as the rider progresses, add resistance to the task, and quickly acquire and solidify this new skill.
Why is it important?
Walking is extremely important for health and independence. UniCycle Project allows researchers to better understand how people acquire alternative complex motor skills and how robotic coaches can help.
Existing tools for walking rehabilitation, such as robot exo sales, are usually pre-programmed in motion or require rudimentary input from the patient. Higgins' research is a step towards a model that interacts with patients and accelerates and improves rehabilitation.
This approach focuses on progressive independence. As learners advance, less robotics support is required over time.
“We don't have an algorithm that tells us exactly what robots should do to help you learn, that's a piece of work that's missing,” she said. By simulating a “pair of robot agents – a “coach” guide during learning of “students” gives you insight into the mechanisms of human learning and adapting the technology to your rehabilitation needs. ”
Interdisciplinary collaboration
Higgins is collaborating with Brady DeCouto, an assistant professor at FSU Unspencer Daybells Education, Health, and Human Sciencespeople who study human learning processes independent of robot involvement. They will work with Shrayas Kousik, an assistant professor at Georgia Tech, with machine learning authority for robotics and reinforcement learning for artificial intelligence systems.
