Localizing objects during robot SLAM in semi-dynamic environments

Abstract

Mapping and localization play important roles for autonomous mobile robots. Since most of the conventional mapping methods assume static environment, the obtained map lacks reliability of localization when the assumption does not hold in the real environment. Additionally, when a robot wants to do some tasks (such as, open doors, push cabinets, etc.) during SLAM process, the robot need localize the target objects. This paper deals with mapping and localization for mobile robot in environments including semi-dynamic objects which change their poses occasionally, such as mobile file cabinets, chairs, and doors. We assume the semi-dynamic objects are target objects for robot tasks.We assume RFID tags with unique ubiquitous identification code (ucode) are attached the objects in place of barcodes currently used. So the tag information allows us to recognize existence of semi-dynamic objects in the environment when a mobile robot detects the RFID tags. To perform mapping and localization for mobile robot, we propose SLAM-SD, an extended SLAM method to cope with the semi-dynamic environments. The method employs a framework of Dynamic Bayesian Network and RBPF (Rao-Blackwellised Particle Filter) which allows simultaneous localization of a robot, mapping of the environment, and also localization of semi-dynamic objects. The system update the occupancy grid map properly when semi-dynamic objects changed the poses. We conducted experiments using a mobile robot mounted a laser range-finder and an RFID tag antenna. The results show effectiveness of the proposed method.