We present an approach for Task-Motion Planning (TMP) that compactly encodes the task-level abstractions within an AND/OR graph growing online. We consider a cluttered table-top scenario where a target object needs to be retrieved. Such scenarios are challenging, since the number of object rearrangements to achieve the target grasp cannot be determined before-hand. However, conventional AND/OR graph based planning requires that the number of re-arrangements are known ahead of time. To address this challenge we propose an AND/OR graph iteratively growing online until the target is retrieved. This iterative graph facilitates faster computations with respect to traditional task planners. We validate our approach using a Baxter robot both in the real-world and in simulation.

https://youtu.be/BXDPMP9MHnk