Mobile Robot Localization Based On Particle Filters In Dynamic Environment

Self-localization is also cited as pose estimation problem. It's a process in which robot gradually gains its estimation of self pose, based on a given map and sensor data. Localization is acknowledged as one the most fundamental abilities for a mobile robot, as a basic function to implement other tasks. However, the noise in sensor data, the local similarities of sensing, and the dynamic feature of real environment become the most challenges of mobile robot localization problem.This paper investigates into the active localization problem for a single robot, in dynamic indoor environment. Based on massive reading of research papers and a comparison of current localization and filter methods, it set on particle filter and info filtering method to solve the problem. The main work in this paper is:Firstly, investigated and implemented mobile robot localization method based on particle filter; experimented and discussed important issues of the algorithm: particle number, resampling method, random particles, adaptive particle filtering and dynamic environment localization; proposed a new method to adjust noise parameters in updating apriori distributions, based on the density of particle congregation. The method is proved to enhance localization speed and precision.Secondly, for dynamic environment, based on the idea of outliner rejection, modeling of sensor measurement and an unknown object rejection algorithm are implemented.Thirdly, proposed a semi-active localization method: it randomly generates target position based on a set cycle, to keep robot out of places where high local similarities may exist. Since path planning is the basics of active localization, this paper also investigated mobile robot local path planning problem, and proposed a new method based on virtual side-slip force, which effectively leads robot out of local trap in clustered environment.Algorithms are implemented in simulation, with data collected virtually as well as from real indoor environment. The proposed localization algorithm based on particle filter is illustrated to be effective and robust in solving classic problems as tracking, global localization and kidnapping, in filtering out unknown object measurement under dynamic environment; and the real-time feature and effectiveness of local path planning method based on virtual side-slip force is also demonstrated.