Bilkent University
Department of Computer Engineering
M.S.THESIS PRESENTATION
MO-SeGMan: A Rearrangement Planning Framework for Multi-Objective Sequential and Guided Manipulation in Constrained Environments
Cankut Bora Tuncer
Master Student
(Supervisor: Asst. Prof. Özgür S. Öğüz)
Computer Engineering Department
Bilkent University
Abstract: MO-SeGMan is a planning framework for multi-objective sequential and guided manipulation in cluttered rearrangement problems with significant task-space constraints. The framework is designed to improve both planning reliability and execution quality by computing object placement orders that account for replanning effort and the robot's travel cost. Essential precedence relations between objects are preserved and utilized to obtain an optimized object placement sequence that minimizes the robot travel distance. For dense, non-monotone scenes, MO-SeGMan introduces Selective Guided Forward Search (SGFS), which focuses relocation actions on critical obstacles that contribute to the current planning failure and moves them to feasible buffer regions. The framework also uses an adaptive subgoal selection and refinement strategy to decompose the difficult manipulation task into feasible intermediate steps. Experiments on 26 rearrangement benchmarks show that MO-SeGMan reliably produces feasible plans and improves over baseline methods in computation time and solution quality. These results indicate that the proposed approach is robust, efficient, and scalable for constrained rearrangement planning.
DATE: June 30, Tuesday @ 10:00
Place: EA 409