Where to begin climbing? Computing start-of-stair position for robotic platforms

Abstract

Multi-floor autonomous robot navigation involves climbing staircases and exploring all the floors in a building using the onboard sensors. While some works investigate the method of exploring indoor floors, other works address the issues in staircase navigation. The proposed work bridges these tasks by allowing a tracked ground robot exploring a floor to find the position where it should begin its climb on the identified staircase. Ideally, such a position is before the first step on the mid line of a staircase, with sufficient space on the floor for the robot to fit in. Various geometric considerations, statistical measurements and randomized algorithms are combined to compute the start-of-stair position from point cloud data obtained from an RGBD camera. An efficient slope finding method is developed to detect the slope containing the staircase. The stair-center on the plane is then computed. The stair bottom, and eventually the start-of-stair position is computed based on the intersection of stair and ground planes. Results are presented on single RGBD frames as well as accumulated point cloud obtained from a mapping algorithm. The scope of error in the initial detection is also analyzed and sample results indicate that the point cloud processing can overcome spatial errors in the initial detection over a considerable margin.