Skip to main content

Available Thesis Projects

Master's thesis on Physics-Informed Neural Network for Simultaneous Muscle Parameter Identification and Joint Torque Prediction

This master's thesis aims to investigate the application of Physics-Informed Neural Networks (PINNs) for the concurrent estimation of muscle parameters and prediction of joint torques in biomechanical systems. The accurate characterization of muscle properties and the prediction of joint torques are crucial in fields such as biomechanics, robotics, and sports science. Traditional approaches often require extensive data collection and complex mathematical models, which can be computationally expensive and lack robustness. In contrast, PINNs leverage both data-driven neural networks and physics-based constraints to efficiently and accurately estimate muscle parameters while predicting joint torques. This research will contribute to the development of a novel framework that combines the strengths of deep learning and biomechanical modeling, offering a promising approach for enhanced understanding and control of human movement and biomechanical systems.

The student candidate is expected to expand and improve the current implementation of the PINN and Hill-type muscle model. To successfully achieve this goal, the candidate should have basic knowledge of neural networks as well as programming skills in Python and TensorFlow (Jax is a plus).



Measuring motion with smartphones
Marker-based motion capture systems are the gold standard for analyzing human motion. Reflective markers are placed in anatomical body segments, and multiple high-speed cameras equipped with infrared light capture the three-dimensional markers’ position. Then, in data post-processing, a human skeleton is reconstructed, providing an accurate description of human motion. This process can be long and tedious. System calibration, marker placement, and data post-processing might require hours. Marker-based systems are also expensive, require technical expertise, and primarily confined to laboratory settings, limiting accessibility and applicability. Markerless motion capture systems that rely on computer vision are promising technologies to evaluate motion quickly and without specialized equipment. Numerous markless capture systems based on RGB and on depth-sensing cameras have emerged in recent years, yet their validation is still limited or scarce. For the reasons mentioned above, this master thesis proposed to evaluate the accuracy of tablet/smartphone based motion capture compared to the golden standard with a full 3D motion lab. The goal of this project is to evaluate the accuracy and validitiy of some newly emerging technology to measure motion with smartphones/tablets. Tasks include to collect data from various movements, such as walking at various speeds, jogging, jumping, etc., in various modalities - a) one tablet with a depth camera, b) two tablets/smartphones, and c) full marker-based 3D movement analysis in the KTH MoveAbility Lab. Then, to compare the computed joint angles and moments obtained from markerless and marker-based systems. This study will provide a better understanding of the capabilities and limitations of current motion capture technologies, as well as insights into human movements across various activities. 



Analysis of balance and balance disturbance

Normal standing requires function from numerous body systems, including the neuromusculoskeletal, somatosensory, and vestibular systems, and as such, balance disturbance can be associated with different underlying conditions. In this project, we focus primarily on orthostatic tremor, a rare condition characterized by high frequency in the leg muscles and a feeling of unsteadiness.

Balance has frequently been explored using posturography, a method that involves analyzing time series trajectory of the center of pressure (COP), the centroid of the summed pressure at the base of support. As standing balance is dynamic, the COP time series can be analyzed in both spatial (path length, range, and velocity) and frequency domains, though there remains a gap in connecting them to the underlying physiological mechanisms. It has been suggested by Zatsiorsky and Duarte that the COP trajectory is made up of a low-frequency component “rambling” – movement of an instantaneous equilibrium reference point that represents the slower postural fluctuations of the reference point, and the high-frequency component “trembling” –COP fluctuations that represent faster postural deviations of the body from its reference point.

The aim of this project will be to study and characterize standing balance in a patient group with orthostatic tremor, during standing trials on a stable surface and on an unstable surface, with eyes open and eyes closed, and to compare it to age-matched adults. The project is performed in collaboration with Danderyd University Hospital.



Belongs to: MoveAbility Lab
Last changed: Mar 10, 2024