Research On Key Algorithms Of Mobile Robot's Perception And Autonomic Planning In Unknown Environment

The perception of the environment and the autonomous planning performance of the mobile robot depend on the accuracy of the self-positioning estimation and the effective map construction.Both of these aspects rely on the data acquisition and data fusion of the sensor.However,the localization and map reconstruction algorithms need to be designed differently for different sensors and dynamic models of mobile robots.The underlying drive and positioning algorithms which based on low-level sensors constitute the "cerebellum" of mobile robots,and the high-level-sensor-based perception algorithm constitutes Mobile robot's "brain".In this paper,corresponding mobile robot platforms for different mobile robots' ‘cerebellum' and ‘brain' are designed for which we set up corresponding dynamic models.Based on the platforms,we studied and verified the related perception and autonomous planning algorithm including location perception algorithms based on low cost ultrasonic ranging sensors,Simultaneous Localization and Mapping(SLAM)algorithms based on laser radar with higher precision and good directivity,and visual SLAM algorithms based on monocular camera.The main tasks of this article are:(1)The architecture design pattern of mobile robots is studied.The perception and self-planning algorithms of mobile robots in unknown environments are divided into bottom drive control and perception and decision modules,which correspond to the "cerebellum" and "brain" of mobile robots.At the same time,two kinds of mobile robot platforms,ROBOT_A and ROBOT_B,were designed and equipped with relevant sensors.(2)In the case that the working environment of the mobile robot is unknown,it is studied to complement the filter by using the local information to establish a good measurement model during the movement of the robot.Therefore,based on the ultrasonic sensor and odometer,an adjacent environmental error model is established and combined with extended Kalman filter and unscented Kalman filter,a positioning and path planning algorithm is designed.After that,we conducted an experiment on the ROBOT_A platform to verify the good positioning accuracy of the algorithm.(3)The SLAM algorithm based on lidar is studied.Based on ROBOT_B equipped with laser radar,data of laser radar data and odometer data are fused,and particle filters are used to estimate the pose of the mobile robot in real time.In the design of the algorithm,this paper introduces the quantitative evaluation particle resampling mechanism,which reduces the unnecessary re-sampling steps in the particle filter to reduce the complexity of the algorithm and greatly reduces the risk of particle exhaustion.In the experiment,we proved the feasibility and effectiveness of the algorithm.In different indoor environments,the algorithm can generate a more accurate grid map with fewer particles per frame under the condition of limited computational power.(4)A feature point matching screening method,Grid-based Motion Statistics(GMS)was studied,and the feasibility of its feature point matching screening engine in monocular visual SLAM was evaluated.Afterwards,the good performance of the GMS method in feature point matching and screening is integrated into the ORB-SLAM system architecture to optimize trajectory tracking and closed-loop detection in monocular vision SLAM by enhancing the performance of feature point matching between frame images.Finally,the monocular visual SLAM algorithm with GMS engine is verified on the ROBOT_A platform.The experimental results show that the improved algorithm inherits the advantages of the ORB-SLAM algorithm,improves the tracking performance,and reduces the misidentification of the closed loop.Rate,but it also inherited the ORB-SLAM flaw.Due to different application scenarios and different costs of different mobile robots in practical applications,this paper aims at different solutions of current mobile robots,combining sensor characteristics,advanced data fusion algorithms and image algorithms,and makes improvements based on the original perception and autonomous planning algorithms.