Authors:

Junyoung Kim

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Object retrieval in cluttered confined spaces is a crucial task in robotics, applicable in fields like surgery, inventory management, maintenance, and cleaning. The task presents significant challenges due to the existence of obstacles and the limited workspace in confined settings. In this work, we aim to find and optimize object retrieval plans via obstacle rearrangement while minimizing the moving distance for the robot end effector. Our approach generates different grasp positions on the target object, figures out the corresponding collision objects, and then leverages the customized Monte Carlo Tree Search (MCTS) to rearrange the obstacles before retrieving the target object. The MCTS rearrangement planner generates robot action sequences to relocate obstacles in the areas that do not block the robot's motion to the target object. This goal-focused MCTS strategy balances exploration and exploitation, which as a result finds short robot action sequences efficiently. We plan to compare our method with the SOTA baselines in the simulation environment before deploying it to the real robot to perform real-world object retrieval tasks. Keywords: Robotics; Object Retrieval; Task and Motion Planning; Artificial Intelligence

Source:

Purdue University / 2024

Topics:

No topics listed

Co-authors:

Junyoung Kim