Robustness Research On SLAM

Simultaneous localization and mapping(SLAM)has been developed for solving many problems in constrained scenes through different sensors since 1980 s.At present,there is a trend of solving problems with environmental adaptability and long-term reliability through common sensors because of the urgent desire in un-constrained scenes.To sum up,the performance is called robustness.For SLAM robustness is devided into two parts,respectively are internal and external factors.This thesis analyzes these two parts of SLAM problems,and improves the overall robustness of the technology and system from three aspects,respectively are system framework,failure detection and accurate processing of complex environment.Firstly,to ensure the compatibility and scalability,this thesis proposes to build a framework around the SLAM system.The safeguard framework based on Robot Operating System 2(ROS 2)is able to provide a good data interface and control interface for SLAM or even other robotic technology.The decentralized Data Distribution Service(DDS)communication system also ensures that the whole system will not crash when any node crash.Secondly,this thesis proposes to improve the runtime stability of the whole sen-sor system from the perspective of internal factors through mutual and independent failure detection.Mutual failure detection is the effectiveness detection of visual sensors and inertial sensors that based on observing each other by two same type robots.Independent failure detection is the long-term monitoring of image quality of visual sensors.In both cases,the abnormal values were judged and intervened.Finally,by the analysis of complex environment,this thesis proposes environment information node,filter node and semantic marking node.The environment information node acquires the information of environment.The filter node analyzes all the data and makes judgments.On the basis of the acquired stable image data and accurate environment information,the moving probability of the readable target in the environment can be judged empirically by the semantic marking node to ensure that the robot is able to estimate more accurately in different environment.It can be proved that the method proposed in this thesis can improve the robust performance of SLAM system in many aspects through unit testing and integration testing of each module.