Abstract
dc:descriptionModeling the impact of fatigue in human motions is essential for developing both animation and biomechanical applications to, respectively, render and explore effects in the kinesiology and pathology of an individual under fatigue conditions. More specifically, in the biomechanics field, fatigue is crucial for creating subject-specific medical strategies and sophisticated ergonomic designs aiming at preventing injuries and chronic fatigue effects. Regarding motion synthesis and graphics frameworks, modeling fatigue in human movements contributes in producing physically-plausible motion sequences for animations and games creating natural action transitions as reactions to external loads causing fatigue effects, as well as for creating realistic digital twins in Virtual Reality (VR), enhancing the immersive experience. Especially, to date, animators, game/VR developers rely on hand-crafted/hard-coded fatigue animations or rendering fatigued motion patterns from motion capture, while their work could be benefited from automatic data-driven solutions. Despite the various advantages such as the above, recent data-driven motion synthesis and musculoskeletal estimation frameworks do not consider the impact of fatigue while generating temporally-evolving motion sequences, allowing for the simulation of a never-ending active human state and the accumulation of non-negligible errors (e.g., not simulating the gradual reduction of maximum exerted joint torques and muscle forces after a repetitive fatiguing task). To address this gap in literature, two fatigue-driven motion synthesis frameworks, namely Fatigue-PINN and FatigueFusion, accompanied by a biomechanical parameter (i.e. joint force) estimation framework enhanced with unsupervised domain adaptation, were developed. Fatigue-driven motion synthesis, falling under the category of "physics-based motion synthesis", requires the manifestation of possible variations in movement patterns and posture due to the inevitable decrease in maximum exerted torques and muscle forces as well as changes in the range of joint angles under fatigue conditions. Hence, leveraging biomechanical laws, 3D musculoskeletal modeling techniques, and physics-informed deep learning methods are required for modeling the temporal dynamics, biomechanical constraints, and subject-specific progression patterns characterizing the fatigue state. To that end, Fatigue-PINN synthetically models the temporal accumulation of fatigue in human movements by exploiting physics-based domain knowledge, while FatigueFusion extends the latter framework by also accounting for the latent infusion of fatigue-specific motion features across time, space and intensity, to produce diversified novel fatigued movements and ultimately biomechanically modeling the whole spectrum of fatigue. Our joint force estimation domain adaptation model complements the above frameworks, by laying the foundation for developing robust data-driven surrogate models for musculoskeletal dynamics estimation tasks, providing insights into the internal body state essential for fatigue modeling, while generalizing to more than one actions classes within the same model. Extensive quantitative experiments across accuracy and variability metrics, statistical analyses, ablation and perceptual studies, and the generation of high quality animation sequences, support the effectiveness and validity of the proposed frameworks. Overall, modeling fatigue in human movements being widely neglected in animation and biomechanics pipelines, as well as fatigue-driven motion synthesis techniques being absent from literature, constitutes the main motivation behind this dissertation.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Loi, Iliana
- Λόη, Ηλιάνα
Subjects
dc:subject × 17- Τρισδιάστατη κίνηση σε υπολογιστή
- Εμβιομηχανική
- Κόπωση
- Βαθιά μάθηση
- 3D computer animation
- Biomechanics
- Fatigue
- Deep learning
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Τεχνολογία μέσων
- Βιοϊατρική Μηχανική
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Media Technology
- Medical Engineering
- Biomedical Engineering
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/62505