Research On Multi-modal Information Fusion For Self-Localization Of RoboCup3D Simulation Robot

With rapid development of intelligent control,computer science,bionics and artificial intelligence,intelligent mobile robot has become the focus in the field of robotics and automation.Autonomous localization is an essential task for intelligence mobile robot navigation,and multi-sensor can collaboratively provide much more information about the robot's state and surrounding.The dissertation mainly focuses on robotic localization algorithm in a known environment and the multi-modal information fusion.Firstly,the techniques used for robot self-localization are reviewed,with attention on the development trend of multi-model information fusion for mobile robot technology.In addition,the related issues of multi-modal information and the corresponding algorithms applied for mobile robot are summarized and discussed.Secondly,a long short memory neural network model is built and can effectively combine the information from the gyroscope,accelerometer,visual,and motion of robot.Once the model is trained,it can predict the robot's position and orientation.We focus on the multi-modal information fusion for robot self-localization based on the long short term memory neural network.Several common multi-modal information fusion algorithms for robot self-localization are also analyzed,such as extended kalman filter and particle filter algorithm.Finally,a series of experiments for robot self-localization are designed based on multimodal information fusion.The experimental results show that the proposed approach makes an improvement in predictive accuracy and effciency compared with the extended kalman filter and particle filter.A profound analysis and discussion of the experimental results is also presented.