| State estimation plays an important role in a mobile robot, since it provides fundamental in-formation for a robot to realize services such as transportation, cleaning, search and rescue, and surveillance. Hence, state estimation is one of the major research topics in the robotics community.Among the field of mobile robot state estimation, the Visual Odometry problem and the Si-multaneous Localization and Mapping (SLAM) problem are two typical problems. In this thesis, we investigate the methods to make the traditional algorithms in these two problems become more robust to the outliers in input data.In the first part of this thesis, we present two efficient approaches to improve the robustness of the RGB-D visual odometry algorithm. First, we consider a new weighted nonlinear least-squares formulation. Different from previous works, we use Expectation Maximization (EM) algorithm to estimate the weights and the motion state alternately. Second, we consider a new bi-objective opti-mization formulation for robust RGB-D visual odometry. We investigate two methods for solving the proposed bi-objective optimization problem:the weighted sum scalarization and the Chebyshev scalarization. Our experimental results for the open source TUM RGB-D dataset show that the new bi-objective optimization formulation is superior to several existing RGB-D odometry methods. In particular, the new formulation yields more accurate motion estimates and is more robust when textural or structural features in the image sequence are lacking.In the second part of this thesis, we present a robust cost function in L1 form to improve the robustness of Linear SLAM. According to the experimental results, such approach contributes very little of value to improve the robustness of Linear SLAM over existing method using L2 form. However, our idea of introducing L1form into least-squares optimization in Linear SLAM offers an alternative beyond the common methods and we need further investigation to improve the idea. |
PDF Full Text Request