Research On Localization Method Of Mobile Robots Based On Multi-sensor Data Fusion

Localization is a fundamental problem for a mobile robot, which is the precondition for the self-navigation operation and attracts wide research attention. To implement localization task, various sensors, such as odometer, ultrasonic sensor, laser radar, visual sensor, GPS and inertial measurement unit (IMU) are developed and different approaches are proposed. However, the traditional approaches with single sensor cannot satisfy the high precision and high reliability demand of self-navigation operation. Multi-sensor data fusion-based approach can avoid the shortage of single sensor and attracts more and more attention. Generally, for multi-sensor fusion-based method, the kinematics model of system and the observation model of sensors are first established, and then position and orientation of robot are estimated based on Bayesian iterative estimation method according to the sensors’measurement and the predictive information of system. Based on the observation of existing sensors and the system model, how to get a higher accuracy result is worth exploring. It is an effective method to improve the estimation accuracy by considering various constraints of the system from the perspective of optimization estimation. However, there are few relative studies in the field of robot localization. In this thesis, some constraints during robot moving are analyzed and an optimized fusion solution for robot localization which takes the environment constraints into consideration is probed.To eliminate the accumulative error of odometer (ODO), based on Extended Kalman Filter (EKF) framework, a fusing method combining odometer and GNSS is adopted; Observation information from GNSS and predictive state value from odometer are registered first and position and orientation of mobile robot are updated.Aiming at the uncertainty problem caused by system parameters and sensor noise, a constraint-based EKF of GNSS/ODO fusion algorithm is presented, the function of the constraint is likely to introduce a local implicit map in the process of filtering estimation, which improve the pose estimation accuracy significantly. The linear and nonlinear state constraint-based EKF model of GNSS/ODO fusion is deduced. Simulations and experiments in real scene confirm the effectiveness of the proposed algorithm.A testing platform with multi-sensor is designed for validating the presented algorithm. The hardware and software of the testing platform are implemented. The hardware of the platform mainly includes system power supply module, an embedded system based on STM32F407 and a wireless remote controller based on Zigbee. The whole software mainly includes embedded system software based on FreeRTOS and LwIP, and GUI interface in upper computer. The reliability and stability of mobile robot platform are tested in outdoor environment.