Manipulation Planning with Tight Geometric Constraints

Many everyday tasks have tight geometric constraints that make sampling-based planning difficult or some task actions infeasible. In these cases, the widely used symbolic planner-based methods become inefficient because symbolic planning with sampled instances exploits only heuristics on symbolic search but has no proper heuristic on which instance to sample first. This issue not only wastes time generating useless samples, but it also enlarges symbolic planning problems, which significantly reduce overall planning efficiency as iteration continues.

We created a hierarchical planning algorithm based on a reachability tree. To improve the algorithm's speed even further, the results of low-level planning are used to inform the high-level planner, allowing for better high-level planning based on experience.

Contact : Kanghyun Kim (kh11kim@kaist.ac.kr)

Open position

None

Related Publications

2023 ICRA "A Reachability Tree-Based Algorithm for Robot Task and Motion Planning" K Kim, D Park, and M J Kim (Accepted) [IEEE] [arXiv]

Videos & Images

icra2023_video_supp.mp4

ICRA 2023 video

hrt_regrasp3-2022-09-21_23.39.16.mp4

Regrasping Problem

hrt_sorting668-2022-09-21_21.56.57.mp4

Sorting Problem

hrt_blocktower6-2022-09-21_00.36.41.mp4
hrt_hanoi3-2022-09-21_21.58.36.mp4

Hanoi Tower Problem

hrt_nonmono3-2022-09-21_00.55.40.mp4
hrt_nonmono3-2022-09-21_00.55.40.mp4