Research On SLAM Algorithm Based On Multi-sensor Data Fusion In Complex Scenarios

In today's world,the continuous change of computer-related technologies and the continuous updating of artificial intelligence algorithms are promoting the development of robot technology,and its related applications are gradually entering People's Daily life,which is the hot research direction of many scholars.Service robots also have a huge market in logistics,catering and housework.Because these service robots all need to have the most basic mobile ability,how to make mobile robots have better autonomous perception and behavioral decision-making ability has become the focus of current research.The application of SLAM(Simultaneous Localization And Mapping)has been widely promoted since the 1980 s And has spawned many different approaches.Its significance is to ensure that the robot can confirm its position in the current environment when it moves from the current position to a target position in the environment.Localization is a key part of SLAM,and the accuracy of Localization directly determines whether the robot can successfully complete its basic work.Therefore,real-time Localization with high accuracy and high robustness is the basic condition for realizing various applications of mobile robot.Although the current mainstream positioning algorithm can help the robot to achieve effective positioning under normal circumstances,the robot will still encounter problems in the positioning of special situations in the face of complex scenes.For example,when the robot is in the environment with large flow of people,complex and changeable,or too high similarity,it is easy to lose its position and cannot timely update its position.Or the distance deduction distortion forms the accumulation error,causes the pose drift;Even the robot will be maliciously "kidnapped",its posture will mutate,eventually make the robot work task failure.Therefore,when the above situations occur,how to improve the robot's positioning accuracy and efficiency,and quickly recover its actual positioning is the focus of this paper.The method of repositioning to correct a robot's wrong position is often referred to as relocation.Aiming at the positioning and repositioning problems of the above mobile robots,this paper studies the SLAM algorithm that integrates multi-sensor data and reinforcement learning technology under the complex scenarios with dense population or strong dynamic environment.Complex scenarios include: stream,obstacles complex moment of the teaching building hall,environment change in the parking lot,environmental information is too similar to the corridor and so on,puts forward the robot can still have good ability of autonomous navigation of SLAM algorithm,optimize the positioning accuracy,improve the efficiency of positioning,can make it in the real complex environment also can maintain good real-time ability,and can quickly adaptive dynamic environment.The following is a brief list of the main research work of this topic:1)Research this topic the research background of the area,and understand the current SLAM and the research status of mobile robot at home and abroad,analyzed the current this topic research hot spot and the mainstream research direction,probability theory and SLAM technology to the current mainstream of robot research analysis,and study its basic principle and implementation method,not only can find the shortcomings of traditional methods and defect,can strengthen the theoretical basis for this topic research content.2)On the basis of the yolov3 object detection model,an improved method that satisfies the real-time performance and recognition accuracy of mobile robots is proposed,and based on this,the semantic information of recognized objects in the environment is obtained.Use laser radar to perceive laser data in multi-directional environment,and then cluster the obtained laser data to generate laser cluster data,and coordinate the camera calibration system with semantic information corresponding to each laser cluster to generate laser data with semantic labels.Through this data fusion technology,it not only ensures the accuracy of data measurement,but also strengthens the characteristics of the observed environmental information.Finally,the method was verified through actual scene experiments,and the landmark system was constructed based on it.3)Propose a method for repositioning using the position information of landmark objects and the pose deduction model built.It can autonomously identify landmark objects and obtain their pose information,record the deflection angle information of the robot through the IMU,and reversely estimate the current pose of the robot.It is possible to complete pose correction without moving the robot and surrounding environment objects.Finally,through the comparison experiment with other positioning algorithms,the method proposed in this topic is verified and analyzed.4)In the current mainstream SLAM method,the odometer is an important factor that affects the operational efficiency and calculation accuracy of the algorithm.The accuracy of the odometer method to record the amount of exercise data directly determines the accuracy of the prediction model during the positioning process.Therefore,this topic uses reinforcement learning to fuse encoder data,IMU data,laser data,and visual data,and use them as the data source of the odometer method.This allows the robot to learn and choose while running in the current environment,adapt to various dynamic environments,and improve positioning efficiency.The method is applied to robot navigation tasks,and its positioning ability is verified through experiments.