动态环境下移动机器人定位及地图创建-控制工程专业论文.docxVIP

Abstract Robot map building in the dynamical and complex environment has been a hot issue in robotics research domain. When there are dynamic obstacles in the environment, laser sensor data for scanning will have an error match and meanwhile duo to the limitations of sensor robot it will generate a global inconsistent map and lead to robot global localization failure in the dynamical and complex environment. To address the issues, this paper proposes a dynamic obstacle detection and filter dummy data points’ method, the global optimization algorithm of fusing stochastic gradient descent method and the nonlinear least squares, sample-based matching rules extend Monte Carlo algorithm. The main innovations are as follows: First, the dynamic obstacle detection and filter dummy data points’ approach is proposed. Taking out three consecutive observations from the laser sensor scanning data points and combining given error function formula and cumulative function formula to determine whether the obstacles are dynamic obstacles or not. When the obstacle is a dynamical obstacle, the paper by filtering the data points to build grid map. Secondly, due to the limitations of robot’s sensor and the noise in the environment, it will bring the question of global map inconsistency. This paper presents fusing stochastic gradient descent method and nonlinear least squares optimization algorithm to simplify the computational complexity and to avoid the problem of high side iterative. After the optimization, the local map can obtain the global maximum likelihood map. Finally, to achieve global localization for mobile robots in a dynamic and complex environment, this paper presents sample-based matching rules extend Monte Carlo algorithm. The extend Monte Carlo algorithm is based on the traditional Monte Carlo positioning algorithm, but when the particle set is poor during the sampling, this paper changes the traditional Monte Carlo resampling method to fusing perception information and