On-going Projects
Smart rehabilitation robots resolving joint rigidity
Funding Agent: Sunhan Hospital (광주선한병원)
Smart Rehabilitation Robot for Ankylosis Relief
Ankylosis refers to a condition where joints become stiff and movement is impaired, often accompanied by pain. Physical therapy after surgery is essential for relieving ankylosis and returning to daily activities. The goal of this project is to conduct physical therapy using a robot at home, work, and in daily life, without the need for a physical therapist.
Currently, there are rehabilitation devices known as Continuous Passive Motion (CPM) machines. However, these devices are bulky and heavy, limiting their use to specific indoor spaces. Therefore, this project aims to create a lightweight and wearable smart rehabilitation robot. This robot will be portable, allowing rehabilitation therapy to be performed anywhere, whether at home, work, or indoors. To achieve portability, the project will implement a motor attachment and detachment feature, allowing the motor to be attached only when in use. Additionally, a user-friendly and lightweight rehabilitation device (robot hardware) will be designed. Furthermore, development is underway to enable the control of the rehabilitation robot using a smartphone application.
Enhancing the performance of wearable lowerlimb assistive robots based on sensorimotor control theory
Funding Agent: National Rehabilitation Center, Ministry of Health and Welfare of South Korea (보건복지부 국립재활원)
This project aims to develop a controller to control the powered lower knee exoskeleton using FEP theory (Free energy principle). FEP is the theory that unifies motor control/perception/learning by minimizing the free energy. The FEP has several advantages over existing methods. In the case of perception, FEP reduces the effect of noise or delays only considering free energy that depends on observed variables and prior about noise. This allows the system to estimate hidden states or filter noise with asmall delay. In addition, FEP needs a small amount of time compared to conventional optimization methods for generating control input and doesn’t need any trained complex learning model (CNN, RNN, etc) to adapt to the environment in which the agent exists. We are trying to apply the FEP to a knee exoskeleton controller to utilize these advantages. In further research, we will apply an FEP-based controller to various cases.
AI-powered gait rehabilitation exoskeleton robots for hemiplegics
Funding Agent: National Rehabilitation Center, Ministry of Health and Welfare of South Korea (보건복지부 국립재활원)
Our group is leading a project aimed at developing an AI-based ground walking exoskeleton robot for hemiplegic walking rehabilitation. Our primary role is the development of control technology for the effective operation of this innovative exoskeleton. In this project, we are conducting technical research on data mapping with and without the use of a walker, trajectory tracking control based on gait data, and compliance control for rehabilitation of the affected and normal sides of gait. The goal of this project is to develop a controller optimized for the rehabilitation of hemiplegic patients, positively impacting their lives.
Analysis of Recurve Archer Movement Data for Enhancing Performance
Funding Agent: Ministry of Culture, Sports and Tourism (문체부)
This research focuses on developing a module that utilizes advanced data analysis and artificial intelligence-based machine learning to analyze the movements of recurve archer using IMU data and their shooting outcomes. The goal is to provide valuable insights that can enhance athletes' performance by leveraging this analysis. The module serves as a key component of an integrated platform, which facilitates systematic management and scientific evaluation of athletes' training and competition records. Moreover, it includes the validation of the accuracy and reliability of video data based on IMU sensor data to ensure effectiveness of the analysis.
Obstable Avoidance and Path Planning for Robot Cleaners
Funding Agent: LG Electronics
This project addresses the issue of robot vacuum cleaners getting stuck on furniture, interior elements, and other obstacles such as tables and chairs during operation. The primary cause of this problem is that obstacles at certain heights do not get detected by the robot's sensors, including lidar, bumper sensors, and laser sensors. The robot perceives these as flat surfaces, leading to collisions and the robot forcibly pushing against these obstacles.
To resolve this, the project proposes the integration of a new path planning approach along with the Free Energy Principle (FEP). This strategy aims to prevent the vacuum cleaner from forcefully entering areas blocked by furniture or window frames, thus reducing the risk of getting stuck and minimizing damage to the furniture. By accurately estimating the likelihood of entanglement before it occurs, the robot can generate and control its path in a way that avoids these obstacles effectively.
Development and Analysis of a Diagnostic Model for Foot and Ankle Disorders Using Public Medical Imaging and Gait Data
Funding Agent: GIST AI기반 융합인재 양성 지원사업
This project seeks to develop a diagnostic model for foot and ankle disorders by linking anatomical features with gait patterns, using a blend of medical imaging and gait data, including a significant dataset from AI Hub's "Gait Video Data for Foot Disorders and Rehabilitation Progress Evaluation." It employs statistical methods and machine learning to analyze the anatomical structures and gait characteristics of foot and ankle patients. The goal is to identify patterns that can lead to rapid and precise diagnoses by exploring the variance in anatomical structures across different disorders. The comprehensive dataset facilitates a detailed examination of foot and ankle anatomy and movement, aiming to overcome previous research limitations related to small sample sizes. Through advanced preprocessing and the application of statistical shape models to X-ray data, the project visualizes bone shapes and orientations, offering an intuitive understanding of disorder-specific differences. The diagnostic model leverages various machine learning techniques to isolate significant variables for disease classification, enhancing clinical insights into foot and ankle disorders.