Inference and Task Planning over Spatially Complex Problems

Abstract

One core problem of robot viability in many sectors is retrainability; if a robot’s task can change without changing code, automation becomes feasible for a wider set of applications. To advance robot retrainability, this thesis will introduce a learning from demonstrations (LfD) framework allowing a robot to learn and execute tasklevel plans in spatially complex environments. To achieve this goal, we introduce a propositional logic framework to encode spatial relationships between objects and an inference scheme to identify important relationships between defined object classes. Finally, we present a search-based algorithm to synthesize required class relationships into a task-level plan. As a representative problem for this of context, we focus on the problem of box packing, wherein the robot must learn specific rules surrounding how to place objects in a box according to a demonstrator’s wishes.

DISTRIBUTION STATEMENT A. Approved for public release. Distribution is unlimited.

This material is based upon work supported by the Under Secretary of Defense for Research and Engineering under Air Force Contract No. FA8702-15-D-0001. Any opinions, findings, conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Under Secretary of Defense for Research and Engineering.