Research On Self-localization Of Indoor Mobile Robot

The research level and the application of robotics reveal the development of the industrial automation of a country, thus they are significant to the national defense and strategy. Mobile robot self-localization is the problem of estimating the robot's pose in the environment based on the self-carried sensors. It is deemed as the fundamental function for the robot to perform other tasks. This dissertation researches on the theory and implementation of the self-localization algorithm of one single robot in common indoor environment based on the odometer and laser range finder.Firstly, this paper describes the problem of self-localization of mobile robot in indoor environment, and systematically introduces the Bayesian Filtering theory as well as its implementations such as Kalman Filter, Extended Kalman Filter and Particle Filter.Secondly, it describes the dead reckoning localization method and probabilistic motion model based on odometer, studies the scan matching localization method based on laser range finder, and analyses three different ways of representing the environment. Since the iterative closest point laser scan matching algorithm requires a good initial registration of two scans, this paper presents the genetic iterative closest point algorithm to deal with this limitation. The new algorithm utilizes genetic algorithm to search the best scan matching solution and thus effectively solve the arbitrary scan matching problem and improve the precise of localization.Finally, this paper studies and implements the self-localization algorithm of indoor mobile robot based on the particle filtering method. It analyses in detail the important factors that influent the performance of the algorithm, such as the proposal distribution, the way to calculate the particle weight, the method of resampling, the number of particle, etc. The global stochastic particles and auxiliary particle filter are introduced in order to handle the robot kidnap problem and localization failure problem. The localization algorithm is further optimized by employing Hill Climbing algorithm to adjust the distribution of the global particles. As a result, the number of particle required and the calculation time are reduced, thus, the efficiency of global localization is improved. Besides, this paper analyses the pattern of scans in dynamic environment where unexpected obstacles show up. Based on the analysis, an algorithm for detecting and filtering the dynamic obstacles in the environment is implemented and the problem of localization in dynamic environment is solved. Moreover, based on the analysis of the calculation time of each operation in the algorithm, the particles in the algorithm are designed to be sampled by stages, so as to make the proposed localization algorithm practical in real time. The various proposed improved method are integrated and a flow chart of the localization algorithm is given. Simulated experiments as well as real world experiments demonstrate that the proposed localization algorithm is effective and robust.