An Object-Oriented Semantic SLAM System towards Dynamic Environments for Mobile Manipulation

摘要

For practical applications of mobile manipulation, the robot should have the ability to model the environment geometrically and analyze the object of interest. If there are people or other robots working around, then the robot also needs to have good adaptability to the dynamic environment. In this paper, we propose a SLAM system which can perform object-oriented semantic mapping of a dynamic environment adaptively. The SLAM system consists of six threads, which can eliminate the effects of dynamic objects, analyze models of previously unseen object instances and create a global three-dimensional map of the environment simultaneously. The proposed SLAM system is verified on the TUM RGB-D dataset. Furthermore, the system is applied in a mobile manipulator for object grasping applications and achieves satisfactory results.