Youngmok Yub in the lab working on his exoskeleton for a hand.
Assistant Professor Ashish Deshpande, a robotics professor in the Department of Mechanical Engineering, reports that his graduate student Youngmok Yun won the UT Graduate School Named Continuing Fellowship. It is one of the most prestigious fellowships offered at the university to a current graduate student. Awards decisions are based on major accomplishments since entering Graduate School, a well-defined program of research, strong personal statement, and letters of recommendation. The fellowship provides one year of support with a $26K stipend, medical insurance, and full tuition and fees.
Excerpts from Dr. Ashish Deshpande's nomination letter:
Dr. Ashish Deshpande
Youngmok joined my lab, the ReNeu Robotics Lab, as a graduate research assistant in Fall 2012. Youngmok is diligent, curious and motivated, and his goal is to develop robotics and controls technologies to assist patients suffering from neurological disorders including stroke.
Youngmok single-handedly developed a statistical modeling technique to analyze human walking patterns based on the motion data collected from 118 healthy adults. He was involved in the data collection phase as a researcher back in Korea. When he arrived here to work with me, he convinced me to pursue this research path in collaboration with researchers from Korea. As a result, we have developed a mapping tool that predicts an arbitrary subject's walking kinematics using their basic body parameters as input. This is a substantial contribution to the fields of human biomechanics and statistical modeling, and I am confident that over the coming years this tool will be heavily used by clinicians and other researchers.
Currently he is working on the design and development of a hand exoskeleton (a wearable robot) for assisting stroke patients during rehabilitation. Youngmok has worked in a team of graduate and undergraduate students in the lab to come up with novel design and controllers for this complex robotic system. Specifically, he is designing an actuation system for the exoskeleton, developing statistical models to predict finger pose based on motion capture data, and developing controllers to achieve reliable movement and force control of the exoskeleton.
Youngmok's description of his work:
How a Robotic's Dad analyzes his Baby's Movement
As a new father, I see many exciting things from my one-year old baby, Chloe. And as a robotics engineer, I observe some of these behaviors from the engineering perspective. When she was born, she did not have any model of her biomechanical system, and the controller in her brain was immature. Despite this, her biomechanical systems were amazingly stable. Over the past year, her ability to move her hand has dramatically improved. Now, she has learned how to manipulate various objects, and she controls them efficiently. As shown in my baby's case, the human biomechanical system has many attractive advantages that are required for the advance of robotics.
My Research Goal
The goal of my research is to derive inspiration from the human control system, and apply its principles to robotics. It will make robots safe, adaptive and versatile in various robotics applications.
A Wearable Robot
I am designing a wearable robot inspired by the biomechanics of the human body. One key feature of human neuromuscular system is elasticity. Soft skin, flexible ligaments, and compliant tendons make the human biomechanical system stable and robust against external disturbance. I studied these biomechanical properties, and applied in the design of a hand exoskeleton for rehabilitation and tele-manipulation (supported by NSF and NASA). The elasticity, inspired by human biomechanics, is introduced to the hand exoskeleton in the form of a novel actuator, called a series elastic actuator (SEA). This made the robotic device comfortable, robust, and safe, which cannot be achieved with conventional robotic structures.
The Hand Exoskeleton Project
One major application of the hand exoskeleton is to assist in the rehabilitation of subjects suffering from stroke, spinal cord injuries and other disabilities. The robot can quantitatively evaluate the progress of rehabilitation by estimating the finger status [1], and generate an optimized rehabilitation motion tailored for each patient. In addition, it may enable doctors and therapists to provide novel rehabilitation interventions.
Mimicking Human Learning Ability
I am developing a statistical modeling method for the robot's "learning" ability. A typical robot controller uses an analytical equation that is provided by the control designer. Therefore, conventional robots can control only specific objects whose analytical equations are already given. Humans, in contrast, are constantly learning the models of their biomechanics and external environments. This learning skill makes humans adaptive to different conditions. For instance, a person can adaptively drive different cars without previous knowledge.
To mimic the human learning ability, I am applying a statistical method to robots. In this method, a robot tries different behaviors (just like a baby does), and collects the resulting input-output data. Then, based on the data, it generates a statistical model, i.e. a regression model. With this approach, I have succeeded in learning:
- the gait motion of human,
- the 3D structure of a building,
- the behavior of a flexible manipulator inspired by a human finger.
I developed a novel control algorithm in order to control a robot that learns its own statistical model. A statistical method has merits in learning the behavior of an arbitrary nonlinear system autonomously, but it also has disadvantages. One critical issue is that the statistical method can generate a reliable prediction only when the data from similar experience has been provided in modeling. Due to this reason, it is almost impossible to directly apply conventional control algorithms for robots whose models are learned with statistical methods.
I recently devised a novel control algorithm to overcome these issues, called "Control in Reliable Region of Statistical Model using Gaussian Process." This control algorithm drives the robot away from such an unreliable region identified with previous data, while pursuing the desired motion by taking advantage of the redundancy in the input-output relationships. For validation, a flexible finger was controlled to demonstrate the practical effectiveness. This method is the first algorithm to control highly nonlinear robotic systems, while addressing the reliability issue of a statistical model.
Long-term Goal- Integration of Men and Machines
My long-term goal is to bring robots into the close proximity of humans. Over the last few decades, robots have worked in structured and confined spaces, such as a factory, isolated from the humans. Robots were too dangerous for humans, and the human environment was too unpredictable for the robots. Now with advances in robot technology, including my contributions in design, modeling and controls, we can explore the potentials obtained from physical cooperation between humans and robots. In the future, I believe that robots will help to overcome the physical limitations of humans in the forms of exoskeletons, prostheses, and autonomous vehicles. Furthermore, future robots will complement human's intellectual abilities with their tremendous computational power. I am confident that with my ongoing research and future research plans, I will continue to make meaningful contributions to this progress.
