Perception & Representation
This pillar addresses how robots can transform raw tactile and visuo-haptic signals into meaningful representations of contact, shape, compliance, and interaction dynamics. I am interested in perception methods that help robots understand not only what they touch, but also how objects and surfaces behave under contact, building toward more robust and transparent interaction in unstructured environments. In the spirit of embodied intelligence, this work connects sensing, representation learning, and task-relevant understanding to support dexterous manipulation and real-world autonomy.
A major open problem is how to build representations that generalize across sensors, tasks, and domains while remaining interpretable enough to support debugging, adaptation, and trust. Current work explores cross-modal learning, latent alignment, sensor transfer, and representations that preserve the structure of physical contact. I am particularly interested in methods that make tactile perception more data-efficient, more robust under distribution shift, and more useful in settings where robots must operate safely and transparently alongside people.
