Sanne van Waveren

Research Abstract:

Safe on-the-fly adaptation is vital to the real-world success of robots that operate with and around humans. Interactive robot learning has shown great potential to teach robots new tasks without extensive programming or robotics expertise. However, the robot’s learned behavior is not necessarily safe or well-perceived by surrounding humans. My research addresses this shortcoming by leveraging techniques from formal methods to ensure safe robot behaviors and improve people’s perception of the robot by integrating human feedback. My work aims at enabling robots to correct their behavior on the fly so they are well-perceived by their users while ensuring desired behavior through specifications. To specify what the robot should and should not do, we can define specifications using a formal language with unambiguous syntax. While experts can provide an initial set of such task and safety specifications, they cannot anticipate all variations in the robot’s environment, the users’ needs, and how people will perceive the robot. For example, an in-home robot tasked with serving a coffee might need to use a cup if a mug is not available, pour less coffee if the user prefers to add milk to their coffee, and keep a comfortable distance from the user when serving the hot coffee. My algorithms endow robots with the ability to query human feedback and refine their task and safety specifications based on the obtained feedback. With my research, I contribute to formalizing human-robot interaction and lifelong robot learning to increase the robot’s autonomy and improve people’s perception of the robot while ensuring safety.

Bio:

Sanne van Waveren is a Postdoctoral Fellow with the CORE Robotics Lab headed by Prof. Matthew Gombolay in the School of Interactive Computing at the Georgia Institute of Technology. Her research enables robots to correct their behavior on the fly so that they are well-perceived by their users. Specifically, she combines techniques from human-robot interaction, machine learning and formal methods to endow robots with the ability to query humans for feedback and correct their behavior specifications based on the obtained feedback. Sanne has organized a series of workshops and seminars to support colleagues to share their research, practice public speaking, and gain experience with academic networking. She has also competed in science slams targeted at a general audience and participated in events to interest young people in programming and robotics. She was selected as an HRI Pioneer in 2022. Prior to joining Georgia Tech, she completed her Ph.D. in Computer Science at KTH Royal Institute of Technology in Sweden.