CAREER: Active Touch Exploration for Understanding Physical Properties

This Faculty Early Career Development (CAREER) project supports research that aims to advance robotic tactile sensing technology by enabling robots to actively explore and perceive the physical properties of objects, such as hardness, texture, and slipperiness. This capability is essential for robots to handle complex tasks, ranging from manipulating delicate fabrics to working with irregular objects like sauces or seasoning particles.

The research looks to develop innovative frameworks that enable robots to co-optimize their actions and observation models, leading to improved perception and manipulation capabilities. This work has significant societal and educational benefits. Enhanced tactile sensing will expand the range of tasks robots can perform in healthcare, manufacturing, and everyday life, making them more versatile and impactful.

Additionally, the project will integrate its findings into new robotics courses, fostering the education of a new generation of roboticists. Openly accessible educational materials will promote broader community engagement and diversity in robotic tactile sensing research, contributing to national prosperity and workforce development.

This CAREER project addresses fundamental challenges in robotic tactile sensing by developing methods for active perception that integrate exploratory actions with tactile signal interpretation. Co-optimization frameworks will be formulated to minimize perception uncertainty by refining both exploratory actions and observation models. These frameworks will draw upon insights from psychological studies, physical modeling, and data-driven optimization, enabling robots to perceive complex object properties.

The research introduces implicit feature representations for properties that are difficult to measure directly, utilizing these representations for both perception and manipulation. This project will provide the first general solution for using active touch to perceive physical properties and for integrating active touch perception with precise manipulation.

By framing the perception-manipulation system as a co-optimization problem, the methods will advance both theoretical understanding and practical applications. A sensorized dexterous hand will validate the approach through challenging tasks involving both solid and formless objects, pushing the boundaries of current robotic perception and manipulation technologies.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.