Real-world data describing how people and vehicles move is often difficult to obtain—it can be scarce, privacy-sensitive, or nonexistent for new settings. My research addresses this by developing models that can generate realistic synthetic trajectories, capturing the variety of ways people and robots actually move. A key challenge is making these models work across large environments while remaining adaptable, so they can handle new settings without needing to be retrained from scratch.
Modeling how things typically move also has a natural counterpart: detecting when they don't. The second strand of my research focuses on anomaly detection for autonomous systems. A robot navigating a busy environment needs to anticipate where others are going, but it also needs to know when its predictions are no longer reliable. I look at how models of expected motion can be used to monitor behavior and flag when something unexpected is happening—helping keep safety-critical systems robust in the real world.
Modeling how things typically move also has a natural counterpart: detecting when they don't. The second strand of my research focuses on anomaly detection for autonomous systems. A robot navigating a busy environment needs to anticipate where others are going, but it also needs to know when its predictions are no longer reliable. I look at how models of expected motion can be used to monitor behavior and flag when something unexpected is happening—helping keep safety-critical systems robust in the real world.
