Research On Autonomous Local Obstacle Avoidance Method Of Sidewalk Dynamic Obstacles For Sweeping Robots

Mobile robot autonomous obstacle avoidance research is a frontier hotspot in the field of mobile robot control.For the sidewalk sweeping robot,autonomous local obstacle avoidance for pedestrians and other sudden obstacles is the prerequisite for efficient cleaning of the sidewalk,and it is also an important guarantee for the sidewalk sweeping robot to achieve autonomous safe and reliable operation.This thesis focuses on the autonomous partial obstacle avoidance method of the sidewalk dynamic obstacles of the sweeping robot,including obstacle information acquisition and obstacle avoidance motion planning of the sidewalk sweeping robot.In terms of mobile robot obstacle information acquisition,compared with other image information acquisition methods,the image semantic segmentation algorithm based on convolutional neural network can achieve pixel-level classification,by dividing and labeling different areas in the image to obtain obstacle information However,when the semantic segmentation algorithm based on AlexNet and ResNet is used to obtain the dynamic obstacle information of the sidewalk,there are problems that the training is not easy to converge and it is difficult to accurately obtain the information of multiple objects in the same segmentation area.In terms of obstacle avoidance motion planning,compared with artificial potential field method,probability graph method and geometric optimization method,the speed obstacle method has good real-time performance,and can eliminate the path jitter problem in the process of dynamic obstacle avoidance,and generate smooth obstacle avoidance The motion planning trajectory,but it is used in the sidewalk sweeping robot obstacle avoidance motion planning has the problem that the robot's own motion information is not considered,resulting in a low accuracy of obstacle avoidance motion planning.Therefore,this thesis studies an obstacle information acquisition method based on VGG16 and Faster RCNN to improve the speed and accuracy of information acquisition;at the same time,it studies an obstacle avoidance motion planning method for a sidewalk sweeping robot based on improved speed obstacle method to improve avoidance The accuracy of obstacle movement planning,which further improves the speed and accuracy of autonomous local obstacle avoidance for sidewalk sweeping robots.The main work completed by the thesis is as follows:(1)A barrier information acquisition method based on VGG16 and Faster RCNN is proposed.In order to solve the problem that the image semantic segmentation algorithm based on AlexNet and ResNet is used to obtain dynamic obstacle information on the sidewalk,the training is not easy to converge,and it is difficult to accurately obtain the information of multiple objects in the same segmentation area.An image semantic segmentation algorithm combined with VGG16 is designed.Pixel-level classification of sky,background and sidewalk in sidewalk image and extract sidewalk boundary line to improve the convergence of the model;further combine Faster RCNN with high speed and accuracy to obtain information of multiple objects in the same segmented area In order to complete the acquisition of obstacle information of the sidewalk sweeping robot,and thus improve the speed and accuracy of autonomous local obstacle avoidance.(2)An obstacle avoidance motion planning method for sidewalk sweeping robot based on improved speed obstacle method is proposed.In order to solve the problem of low accuracy when the existing speed obstacle method that does not consider the robot's own motion information is used in the sidewalk sweeping robot's obstacle avoidance motion planning,the speed obstacle method is implemented by combining the obstacle information and the sidewalk sweeping robot's own motion information.Improvements to improve the accuracy of autonomous local obstacle avoidance.(3)Build an experimental platform for autonomous partial obstacle avoidance system of sidewalk sweeping robot,design a hardware platform to complete the image collection and transmission of sidewalk sweeping robot,and use Matlab to build and train the network model,and realize the autonomous partial obstacle avoidance system software for sidewalk sweeping robot Developed,and finally completed the autonomous partial obstacle avoidance experiment of the sidewalk sweeping robot based on the experimental platform.The experimental results show that,compared with the AlexNet and ResNet image semantic segmentation algorithms,using the proposed obstacle information acquisition method based on VGG16 and Faster RCNN,the average class accuracy of semantic segmentation is increased by 4.58%,and the average crossover ratio is increased by 4.52% The weighted cross-combination ratio has increased by 3.46%,the training time has been reduced by 21.8%,the information acquisition time has been reduced by 9.5%,and the accuracy rate has increased by 2.77%.Based on the obtained obstacle information,the speed obstacle method before and after the improvement is used for obstacle avoidance motion planning.Compared with the existing speed obstacle method that does not consider the robot's own motion information,the accuracy of the obstacle avoidance method using the improved speed obstacle method is improved.5.1%.Compared with the autonomous local obstacle avoidance method based on AlexNet,ResNet and the existing speed obstacle method,the proposed local automatic obstacle avoidance method for the sidewalk dynamic obstacles of the sweeping robot reduces the time for obstacle avoidance by 5.69%,and the accuracy of obstacle avoidance is improved.7.54%.Experiments verify the effectiveness and superiority of the autonomous partial obstacle avoidance method proposed by the sidewalk sweeping robot.