Case-Based Object Placement Planning

Kellen Gillespie, Kalyan Moy Gupta, Michael Drinkwater

Abstract: Autonomous object placement is a central task in many military and industrial applications. Deciding where to place an object, or placement planning, is a crucial reasoning task within object placement. Placement planning is especially challenging in 3D worlds. In previous work we developed approaches for placement planning that perform well but require manually-engineered knowledge. Consequently, we introduce case-based object placement (COP) planning to reduce the manual engineering burden and improve portability. COP includes the ability to automatically learn cases from problem solving sessions and reuse them to solve previously unseen placement planning problems. We evaluated COP on object placement tasks in virtual office environments. Our findings show that COP can reduce knowledge engineering and increase portability without compromising placement accuracy.