Recognition of 3D dynamic environments for mobile robot by selective memory intake and release of data from 2D sensors

We have developed an autonomous mobile robot to perform in-hospital transportation. When an autonomous mobile robot moves in a hospital, it needs to detect and avoid such obstacles as beds, wheelchairs, patients, visitors, and hospital staff. These obstacles move and often have overhanging parts that cannot be detected by only 2D sensor which measures horizontal surface. The robot, therefore, must recognize obstacles in a three-dimensional space on a real-time basis. Appropriate and practical 3D sensors for actual use are not available, and 3D recognitions using 2D sensors such as 3D-SLAM are not fast enough to collide with moving object. This paper proposes a method that recognizes the three-dimensional environment around a mobile robot using a laser range finder at its front for measuring the horizontal surface and three additional range finders at the left and right sides and the front top position to look down and detect three-dimensional obstacles. The information is integrated by data fusion into the memory. The features of this method include memorization of the obstacle positions with overhangs and removing the memorized position of the moving obstacles from the memory by selective memory intake and release of data from 2D sensors. This paper also describes experimental and actual operational results in detail about the robot equipped with the developed process in real hospital.