| With the rapid development of science and technology,especially the era of big data and artificial intelligence is gradually approaching,and robots as a product of the era play an increasingly important role.In the process of robot movement,stairs may be the road that the robot must pass Therefore,how to make the robot pass the stairs smoothly and accurately is particularly important.This article focuses on how self-developed multi-wheeled foot robots go through stairs.For the problem of staircase recognition,this paper proposes to train the image classification model to identify the staircase.Compared with the traditional image method,the recognition effect is better and the performance has been greatly improved.For the locating problem of the staircase,this paper determines through the experiment The edge detection algorithm that is most suitable for stairs is used,and then the improved contour extraction algorithm is used to locate the stairs,so as to give the robot a suitable path to advance.For the self-balancing problem of multi-wheeled robots on the stairs,this paper first improves the structure Based on the multi-wheeled obstacle-crossing robot,the sequential machine learning model is designed through the data collected by mpu6050.The traditional control method cannot solve the self-balancing problem in an unstructured environment.The method designed in this paper is to solve the unstructured The self-balancing problem in a globalized environment provides a completely new solution,and in the experimental environment of this paper,the time-series machine learning model designed in this paper can better solve the self-balancing problem of multi-wheeled robots on the stairs.For identifying stairs.This paper comparatively studies the traditional image classification network cnn,resnet network and template matching.Collecting enough data sets for testing,comparing the three methods,the experiment shows that the resnet network has a better effect and can better identify the stairs.Its recognition accuracy rate is 99.3%,and the recall rate is 99%.For the positioning of stairs.In this paper,through the early exploration and analysis,the overall process is determined,that is,image preprocessing,then edge detection,and finally extraction and screening of contours.For edge detection,this paper compares the sobel edge detection operator,canny edge detection algorithm and Laplacian operator.Experiments show that the sobel edge detection operator has good results in various situations,and then through contour extraction and filtering,the final staircase is located.Aiming at the self-balancing problem on the stairs,that is,the self-balancing problem in the non-structural environment.For the improvement of the underlying mechanical structure: This article redesigned the multi-wheeled coupling shaft,flange(connector part)and sleeve,replaced the coupling shaft with a six-wheel structure,and replaced the flange with a matching prismatic and circular arc shape The structure,the sleeve and the corresponding gear link part have also been modified accordingly.For the design of upper-layer control algorithms: This article firstly explores and analyzes through the early stage.Traditional self-balancing control methods cannot solve the problem of self-balancing in unstructured environments.This article cuts from the establishment of machine learning model to solve this problem.Obtain the angle information of the multi-wheeled foot robot when climbing the stairs through mpu5060,in order to design a time series machine learning model.In the model,the traditional svm model and the xgboost model are compared respectively.The experiment shows that as the number of states increases,the accuracy of the xgboost model is getting higher and higher.When 45 gyroscope data is taken to build a machine learning model,the xgboost model is testing.The accuracy rate of 91.16% can be obtained on the set,which is greatly improved compared with the traditional method.In the prototype experiment,it can better solve the self-balancing problem when the multi-wheeled robot climbs the stairs.Based on the optimized design results,build a prototype for verification testing.For the staircase recognition ratio problem,the resnet network model designed in this paper can detect the staircase better and has better adaptability;for the staircase positioning problem,the sobel edge detection operator determined in this paper is used,combined with image preprocessing And contour extraction,this method is good enough to locate stairs.For the self-balancing problem of multi-wheeled robots on stairs,the method based on sequential machine learning model designed in this paper can better solve the self-balancing problem in unstructured environments. |
PDF Full Text Request