Altan Akat

Advisors

Prof. Dr. Anne Koelewijn, Dr.-Ing. Eva Dorschky, Markus Gambietz (M. Sc.)

Duration

05 / 2022 – 11 / 2023

Abstract

This master thesis aims to estimate musculoskeletal states utilizing wearable sensors, spe- cifically inertial measurement units (IMUs). This enables the development interactive systems that can provide feedback or support human motion in natural environments, such as gait retraining, injury prevention, or exoskeleton control. Existing methods proposed in literature include physics-based approaches (e.g., [Dor+19]) and machine learning (e.g., [Mun+21]), but these have limitations. Physics-based approaches are computationally ex- pensive and not ideal for feedback systems, while machine learning requires a representative dataset for training, which can be time-consuming and expensive to collect. Additionally, it is impossible to account for all possible conditions in natural environments.
Prior research has shown that using simulated data can improve the performance of deep neural networks that map IMU data to gait biomechanics [Dor+20], but this approach was limited by simulations being performed independently of the learning process. Thus, this project seeks to combine deep learning and musculoskeletal modeling and simulation to im- prove the generality of neural networks to previously unseen real-world inputs. The project will begin with a literature review on physics-informed machine learning and its application to human motion analysis. The study will investigate methods to embed physical know- ledge into the deep learning framework (e.g., [Hei+21; Rai+19]). Finally, the project will implement and evaluate the embedding of musculoskeletal dynamics into neural networks.

References

[1] [Dor+19] Eva Dorschky et al. “Estimation of gait kinematics and kinetics from inertial sen- sor data using optimal control of musculoskeletal models.” Journal of Biomechanics 95 (Oct. 2019), p. 109278.
[2] [Dor+19] Eva Dorschky et al. “Estimation of gait kinematics and kinetics from inertial sen- sor data using optimal control of musculoskeletal models.” Journal of Biomechanics 95 (Oct. 2019), p. 109278.
[3] [Hei+21] Heiden, Eric et al. “NeuralSim: Augmenting Differentiable Simulators with Neural Networks.” 2021 IEEE International Conference on Robotics and Automation (ICRA) (2021): 9474-9481.
[4] [Lhr+21] Julio S Lora-millan, Andres F Hidalgo, and Eduardo Rocon. “An IMUs-Based Extended Kalman Filter to Estimate Gait Lower Limb Sagittal Kinematics for the Control of Wearable Robotic Devices.” IEEE Access PP (2021), p. 1.
[5] [Mun+21] Marion Mundt et al. “A Comparison of Three Neural Network Approaches for Estimating Joint Angles and Moments from Inertial Measurement Units.” Sensors (Basel, Switzerland) 21.13 (2021).
[6] [Rai+19] Raissi, Maziar, Paris Perdikaris, and George E. Karniadakis. “Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations.” Journal of Computational physics 378 (2019): 686-707.
[7] [Ste+20] Bernd J. Stetter et al. “A Machine Learning and Wearable Sensor Based Approach to Estimate External Knee Flexion and Adduction Moments During Various Loco- motion Activities.” IEEE Journal of Biomedical and Health Informatics 24.2 (2020): 426-434.