Title: User-Adaptive Object Rearrangement Alongside and in Collaboration with People
Date: Monday, June 9th, 2025
Time: 12 PM ET
Location: Klaus 1120A
Virtual link (zoom): https://gatech.zoom.us/j/95599584934
Kartik Ramachandruni
Robotics Ph.D. Student
School of Interactive Computing
Georgia Institute of Technology
Committee:
Dr. Sonia Chernova (Advisor) - School of Interactive Computing, Georgia Institute of Technology
Dr. Harish Ravichandar - School of Interactive Computing, Georgia Institute of Technology
Dr. Zsolt Kira - School of Interactive Computing, Georgia Institute of Technology
Dr. Tesca Fitzgerald - Computer Science Department, Yale University
Dr. Akshara Rai - Meta AI
Abstract:
The object rearrangement problem in robotics is the challenge of manipulating objects to bring the user’s environment to a desired goal state. Effective object rearrangement is characterized by: 1) autonomy – the robot's behavior must be autonomous, requiring no input from the user, 2) personalization – the object arrangement should match the user’s individual organizational style, 3) generalization – the robot’s rearrangement strategy should generalize to novel objects and environments, and 4) user-alignment – if the user is present, the robot's modifications to the environment must align with human activities and actions. To endow the robot with the above characteristics, this thesis introduces planning frameworks for communication-free object rearrangement alongside and in collaboration with people. These frameworks incorporate semantic task information, observed user actions, and prior environment states to learn object rearrangement preferences and task strategies, generalizing to previously unseen objects and environments in a zero-shot manner. Specifically, this thesis presents four key contributions: (1) A human-robot collaboration algorithm that leverages observed human actions and prior task knowledge to actively select robot actions that complement human behavior and maximize task efficiency, thereby enabling mutual adaptation; (2) Object rearrangement frameworks that utilize prior object arrangements, the current environment state, and environment semantics to place objects meaningfully in partially arranged environments; (3) A crowdsourced benchmark and dataset collected from 72 online users performing household organization tasks involving diverse objects and environments, to evaluate personalized object rearrangement in partially arranged settings; and (4) A proposed approach for relocating misplaced objects in human-occupied environments without interrupting ongoing user activities.