Publications

Collaborative robots must keep adapting to new tasks and user preferences without overburdening the person they work with. We cast multi-task interactive learning as a facility location problem, planning a sequence of human interventions that is near-optimal across a whole stream of tasks rather than greedy on the current one.

Shivam Vats, Michelle Zhao, Patrick Callaghan, Mingxi Jia, Maxim Likhachev, Oliver Kroemer, George Konidaris

Robotics: Science and Systems (RSS), 2025

Uncertainty quantification lets a robot ask for help when it leaves familiar territory — but standard conformal prediction assumes exchangeable data, which interactive imitation learning violates on two fronts. We give a calibration method that stays statistically valid when the expert’s own behavior shifts and when feedback arrives only intermittently.

Michelle Zhao, Reid Simmons, Henny Admoni, Aaditya Ramdas, Andrea Bajcsy

International Conference on Learning Representations (ICLR), 2025

Assistive arms have far more degrees of freedom than a joystick can express, so learned interfaces must guess at the operator’s intent. We calibrate that mapping with conformal prediction, giving the robot a principled signal for when its inferred action is trustworthy and when it is not.

Michelle Zhao, Reid Simmons, Henny Admoni, Andrea Bajcsy

Robotics: Science and Systems (RSS), 2025

People care not only about whether a task gets done but about which parts of it they do themselves. We model and learn these contribution preferences, so a robot can divide collaborative work in the way its partner actually wants.

Michelle D. Zhao, Reid Simmons, Henny Admoni

Conference on Robot Learning (CoRL), 2023

For people to coordinate with robots, they need an accurate picture of how the robot will act. We generate explanations of a joint multi-agent strategy and show they improve how well people can predict and work with a robot partner.

Ravi Pandya, Michelle Zhao, Changliu Liu, Reid Simmons, Henny Admoni

IEEE International Conference on Robotics and Automation (ICRA), 2024

Rather than expecting people to adapt to the robot, we infer a partner’s high-level strategy from a handful of observed actions and adapt the robot’s policy to match it.

Michelle Zhao, Reid Simmons, Henny Admoni

IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2022