Active Path Clearing Navigation through Environment Reconfiguration in Presence of Movable Obstacles

摘要

Geometrical map based path planning is a well adopted approach for robot navigation tasks. Although straightforward, it can only deal with the environment changes in a reactive way (passive re-planning) and cannot handle planning failures. In this paper, we try to tackle this problem with a smarter approach - using local environment reconfiguration strategy to actively create clearances through manipulation of movable obstacles. Basically, we implement our previously developed deep Convolutional Neural Network (CNN) based perception module to update local environment knowledge and conduct online space reconfiguration planning for local path clearing. We develop a novel path clearing algorithm capable of dealing with ordered manipulations of multiple movable obstacles to provide locally optimal navigation solution. We illustrate the pipeline implementation of the overall system and verify the effectiveness with simulations as well as real-world scenarios, regarding different objects and manipulation actions.