Research On Robot Simultaneous Localization And Mapping

Simultaneous Localization and Mapping (SLAM), an important branch on the mobile robot system, is to ensure whether robot exploration can be completed independently in unknown environments.Mapping, map merging and cooperative localization, the key issues in robot SLAM, have been studied deeply, and some solving schemes are suggested. The research work can be summarized in the following aspects:(1) Based on detailed analyses of uncertainty sensor information processing in ultrasonic ranging, especially on the scattering and specular reflection, the novel methods of mapping building are presented. In probabilistic grid maps, range confidence factor is introduced in ultrasonic model, and improves the map accuracy. In feature maps, a novel method is advanced based on Randomized Hough Transform (RHT) and Multi-resolution Hough Transforms (MHT). The method regards all discrete points as seed points, and randomly picks a part of discrete points for pairing with seed points. Corresponding accumulator cells are incremented in the space. When the accumulator cells reach the maximum, the line is extracted. So uncertainty ultrasonic sensor responses can be successfully reduced and the robustness for measuring distance is proved.(2) The methods of decentralized exploration and concentrated mapping are adopted in the robot map building system. Firstly, each robot starts from different positions and builds the local map; secondly, the improved differential evolution (DE) algorithm is used to search the transform functions; thirdly, according to the transform functions, the map is rotated and translated so that the overlapping region is maximum and the dissimilarity is minimum; finally, accepting function determines whether map merging is completed successfully. The improved different algorithms adopt a novel mutation which combines mutation strategy DE/best/1with DE/rand/1by a linear simulated annealing strategy, and linear annealing factors are used as weighting factors, which improve the convergence precision and convergence speed in local maps merging. To examine the Hough transform based map merging algorithms, we research on the algorithm’s performance in a series of elemental tests, such as overlapping and rotation. We focus on the performance when only subsets of points are picked randomly or in a deterministic way to compute the transformation.(3) Via a deep study on SLAM, this paper introduces a novel method of SLAM based on particle filters to solve particle impoverishment and particle depletion. The solution to the first problem is to integrate particle swarm optimization (PSO) with FastSLAM. Through the particle swarm optimization, the predictions of particles are updated, and the particle’s proposal distribution is adjusted, so the accuracy of position prediction is enhanced. To solve the second problem, the mutation operation based on genetic algorithm is adopted in PSO, so as to keep the particle diversity. Then, the PSO is applied to FastSLAM algorithm for heterogeneous multi-robot. In heterogeneous multi-robot system, we make full use of the robot ability of accuracy localization. The relative locations (including distance and orientation) are calculated between the robots, so the predictions of particles are updated considering the relative observations. By so doing the precision of localization is improved.(4) Based on sparse extended information filter (SEIF), a new sparse rule is proposed. In the sparse rules, active sets and the sets with active features that we seek to deactivate are decomposed into two parts:the overt and covert. Then the associations between the robot and the features are adjusted according to the observations of the next moment, and the elements of two sets are changed corresponding, thus ensuring the removal of a weak connection and retain a strong connection. As to the questions of data association, incremental data association methods based on Mahalanobis distance are adopted, which have improved the matching accuracy, reduced the error data of map merging, and achieved robot SLAM in a large-scale environment.