Multi-sensor System Design And Its Application In Robot Localization

In RoboCup Match, the traditional and common approach to self-localization for a robot is to identify some marks in the field (say the goal and sideline) based on an omni-directional vision system. Actually, the developing trend of middle size league indicates that Localization just by vision faces more and more challenge. So, the method based on information fusion with Adopting much more other types of sensors is a new and prevail research area for self-localization of soccer robot.This thesis introduces the design and implement of a multi-sensor system, which is composed of encoders, accelerations and a compass. Firstly, after describing the developing history and systemic characteristic of multi-sensor systems, this thesis sums up several fusion structure and analyses some concerned points of multi-sensor systems. In this thesis, the robot's multi-sensor system is delaminated to three layers and separated into several models, and the design and implementation of these models'function is discussed and analyzed in detail. Then, the conception of smart sensor is introduced for multi-sensor system design to improve systemic characteristic including: flexibility, universality, extendability, extendability, data preprocessing, self-adjustability, data transmission by BUS, and so on. According to this, a universal hardware platform is designed for various types of sensors by adopting microcontroller PIC18F258 and an embedded real-time operation system (RTOS)——PICos18.A 4-wheel omni-directional robot is used for RoboCup MSL, which is much different from traditional two-wheel mobile robots in kinematic characteristics. This thesis presents an odometry model for 4-wheel omnidirectional mobile robots and proposes a method to achieve holonomic description of robot's kinematic characteristics based on two dual-axis accelerations. With combinated data from encoder,acceleration,compass and pose from an omni-directional vision system, a Kalman Filter is designed to fuse data for robot's self-localization. Results of experiments concludes the achievement of this thesis.