ºÚÁϳԹÏ×ÊÔ´

Structured Intelligence for Robot Manipulation

Contact-rich manipulation requires compliant bodies for safe, adaptive interaction, yet remains challenging because robots must reason about real-world complexity and uncertainty in real time. This talk presents a perspective on building structured intelligence for physical interaction. We begin with compliant and proprioceptive actuation for robotic manipulation, showing how soft-rigid hybrid structures can retain the safety and adaptability of soft materials while gaining the sensing, observability, and control precision needed for dexterous manipulation. We then examine structured tactile sensing and visual-tactile learning as ways to recover useful physical information from complex contact signals. Finally, we turn to robot learning: rather than assuming that more randomized data necessarily produces better policies, we propose the empirical Neural Tangent Kernel analysis to distinguish memorization, adaptation, and invariance to irrelevant variation. Across mechanism design, perception, and learning, we advance structure-conditioned intelligence: robot capabilities are shaped by physical structure in their bodies, their sensory representations, and their learning policies, enabling robust and generalizable manipulation.

Chao Liu is an Assistant Professor in the Department of Mechanical Engineering at The University of British Columbia, Vancouver.