I am a Computer Science Ph.D. student in the Parasol Lab at the University of Illinois Urbana-Champaign, advised by Prof. Nancy M. Amato. I build extended reality (XR) systems that let people specify and teach robot behavior by editing the robot's environment rather than the robot itself, and I develop the planning and learning methods behind them, from preference learning to scalable multi-robot motion planning. I received my B.S. in Computer Science, with a minor in Mathematics, from the University of Wisconsin–Milwaukee.
My work sits where human-robot interaction, motion planning, and robot learning meet. Recently I presented CIPHER at WAFR 2026 and ERUPT at ICAT-EGVE 2026, and I am running a pre-registered user study comparing immersive and desktop interfaces for robot motion planning.
Immersive interfaces for programming, previewing, and supervising robots. Users reshape the robot's workspace and the planner responds in real time.
Learning robot preferences from demonstrations and corrections, including environment edits that disambiguate what a person actually cares about.
Scalable planners that coordinate robot teams through workspace-level guidance, for geometric and kinodynamic robots.
An open-source XR toolkit for interactive motion planning. Users create, scale, and move obstacles with their hands. Edits sync to the MoveIt planning scene in about 40 ms, and planned trajectories can be previewed in VR/AR before running on a physical UR5e. A pre-registered user study comparing ERUPT with the 2D MoveIt/RViz interface is underway.
CodeA hierarchical multi-robot planner that coordinates robots through a workspace decomposition and refines it wherever conflicts arise, so it rarely has to plan in the expensive joint space of all robots. It solved 100% of benchmark problems with up to 32 robots and was up to an order of magnitude faster than state-of-the-art baselines.
Project Page CodeWhen a person drags a robot arm away from a laptop, the robot can't tell which of many changes they meant. FERRET lets people edit the scene instead while the robot's motion stays fixed. It learns preferences from both scene and robot edits using relational and graph neural network models, and recovers features that robot corrections alone cannot.
An XR interface for teaching robots by demonstration with a virtual arm. Users can safely show the high-cost states that are dangerous to demonstrate on real hardware. Features learned from XR demonstrations matched those learned from demonstrations on a physical UR5e.
Paper Code* denotes equal contribution