Phd in DATA SCIENCE

Thesis title: On Modelling Humans: Forecasting, Synthesis and Human-X Interaction

Modeling human motion and interaction is fundamental to advancing technologies in robotics, virtual reality, autonomous systems, and behavioral analysis. The ability to understand and predict human movements and social dynamics opens up new possibilities for machines to interact with people in intuitive and effective ways, contributing to areas such as human-robot collaboration, motion synthesis, and anomaly detection. Research into human dynamics plays a pivotal role in developing systems that can anticipate future actions, generate realistic behaviors, and adapt to unpredictable environments. This work addresses these challenges through a sequence of studies, starting with human pose forecasting using graph convolutional networks (STS-GCN), which models the intricate space-time correlations of body joints to predict future poses. It then extends this framework to collaborative environments, enabling robots to predict human movements in industrial settings (SES-GCN). The exploration continues with the prediction of motion between interacting people (2BODY) and expands further to scene-aware human trajectory and pose forecasting (STAG), focusing on the interaction between humans and their environments. The study of human dynamics also delves into specific applications, such as detecting anomalies in human behavior (COSKAD) and forecasting player trajectories in sports, where role-based interactions within teams play a key role (NBA). The latter part of this research builds upon these foundational insights, proposing methods for egocentric 3D pose estimation in video sequences (SEE-ME) and tackling the generation of human motion sequences from textual descriptions with variable lengths (LADiff). Finally, my research addresses real-time human-robot collaboration, where robots learn to follow humans and adapt to social dynamics to avoid collisions in shared spaces (SDA). Altogether, this body of research highlights the importance of human motion modeling and social interaction, paving the way for intelligent systems capable of seamlessly integrating into human environments and enhancing their ability to interact, collaborate, and coexist with people.