Interaction-Aware Exploration in Cluttered Environments with a Mobile Manipulator

Submitted for review

Exploration with mobile manipulator has received increasing attention from both academia and industry. In many existing works, spaces occluded by objects are typically considered inaccessible. However, for real-world applications like search and rescue, it is crucial to investigate these occluded spaces by interacting with the environments. In this paper, we propose an interaction-aware exploration framework to thoroughly explore cluttered environments containing obstructed spaces with an eye-in-hand mobile manipulator robot. We introduce a new representation called hidden frontier, which can efficiently guide robot to discover occluded unknown space by active interaction. Moreover, a constrained whole-body configuration database is designed to quickly query and output the feasible whole-body configurations given a desired sensor pose during exploration planning. After that, a global path planning is formulated as an Asymmetric Generalized Traveling Salesman Problem (AGTSP) to solve for a minimum-time viewpoints visitation sequence. Simulation experiment results show that proposed method is capable of efficiently exploring the environments more thoroughly and producing voxel maps with higher coverage through active interactions.