Robotics PhD Dissertation Defense Presentation: Steven Hyland
11:00 a.m. to 1:00 p.m.
Active Estimation of Object Properties Through Non-Prehensile Robotic Manipulation
Abstract: Robots often encounter unfamiliar objects whose physical properties are unknown. A camera can reveal an object’s shape and appearance, but cannot reliably determine its dynamics. Through deliberate physical interaction, robots can learn about these properties and use this knowledge to improve their manipulation capabilities.
This work explores how robots can use non-prehensile manipulation, such as pushing, pulling, and tipping, to better understand and interact with unfamiliar objects. By applying controlled forces and measuring the resulting motion, robots can estimate otherwise unknown physical properties without needing to grasp or lift objects. Beyond estimation, this work investigates manipulation strategies for safely repositioning objects under uncertain contact conditions, including tasks that require precise, goal-directed motion. Finally, these interactions provide a foundation for generating physics-based simulation models, bridging the gap between real-world objects and their digital counterparts.
Advisors: Professor Jing Xiao and Professor Cagdas Onal
Committee: Professor Constantinos Chamzas, Professor Ziming Zhang
Zoom Link: https://wpi.zoom.us/j/8173956045