Research On Simultaneous Localization And Map-building Of Mobile Robots

With the development of mobile robot in research, Simultaneous localization andmap-building(SLAM) has attracted immense attention in mobile robotics field. The SLAMproblem has attracted a lot of researchers with a broad range of interests and application.This dissertation focuses on solving SLAM problem in unknown outdoor environment.Several improved methods are presented in order to improve consistency and computationalefficiency，and additionally extend SLAM application domains. The main content of thisdissertation include the following aspects:1、Strong tracking filter (STF) is lack of adaptive ability for systems with time-varyingnoises that seriously decreases the estimation accuracy of system state. In order to overcomethe drawback, this paper proposes a fast-denoising adaptive strong tracking filter (FASTF)with the online adaptive estimation for the noise covariance matrixes, the effects of noises onsystem state estimation are suppressed, and the system state estimation converges to realvalues quickly. Performances of STF and FASTF in environments with changing noises arecompared by simulation. The experimental results show that FASTF is of better stateestimation accuracy and adaptability.2、In allusion to the simultaneous localization and mapping(SLAM) problem of mobilerobot in three-dimensional space, a fast and adaptive SLAM algorithm based on improvedstrong tracking filter(STF) is proposed in this paper. Firstly, the noise covariance matrixes ofSTF are adaptively estimated on line, the effects of noises on system state estimation aresuppressed, and the system state estimation converges to real values quickly. Secondly,singular value decomposition (SVD) is used to decompose the state covariance matrix whichcan help the algorithm get better numerical stability. This algorithm can improve the adaptiveability for time-varying systems and the state estimation precision, and the comparativesimulation results with STF-SLAM algorithm illustrate the effectiveness and the superiorityof the proposed algorithm.3、An adaptive landmark detection algorithm. The basis of modeling the environment and scan matching is the landmarks in the environment. In order to extract landrnark points fromlaser range data, we proposes an adaptive entity landlnark detection method.The spider web isintroduced in to separate objects that occupy a range in the environment. The entities whichmeet certain standards are extracted as landmark points.