Task and Motion Planning for grasping targets with object rearrangement in confined workspaces

Abstract

We propose a planning algorithm for rearranging objects with a robot manipulator to grasp a target object within a cluttered environment. Our algorithm accounts for challenges posed by the high density of surrounding objects, which might render some objects undetectable by sensors and make it difficult to find collision-free paths for prehensile manipulation. The main focus is to optimize the efficiency of the overall task by minimizing the number of rearrangement actions to relocate obstacles. To achieve this, our algorithm generates a sequential order of object rearrangement using a combination of prehensile and non-prehensile grasping skills. By leveraging a search-based approach, our algorithm finds a rearrangement plan with the minimum number of actions among multiple solutions. In scenarios involving non-prehensile manipulation, we employ a heuristic function to evaluate the friction between objects and the table, aiding in the decision-making process. Furthermore, our algorithm incorporates the concept of invisible space as a probability calculation, allowing the search method to find currently undetectable objects by sensors. Through extensive experimentation in a real-world setting, we demonstrate the effectiveness of our algorithm in reducing the number of rearrangement actions required, thus showcasing its practicality and performance.