Research On Motion Planning Of Autonomous Excavation Based On Machine Vision

At present,the level of automation of earthwork excavation is relatively low,and the recognition of penetration point and dumping point,excavation trajectory planning,and action execution of excavation operation environment is still completely dependent on the decision analysis and operation skills of skillful operators.To realize autonomous excavation,this dissertation adopts the methods of theoretical analysis,numerical simulation,and test research to systematically and further study the key technologies of visual perception and motion planning based on machine vision technology,such as the 3D reconstruction of the excavation area,the detection and localization of the penetration and the dumping point,the trajectory generation of autonomous excavation movement based on the spline and the spatial-temporal optimization method of excavation trajectory.The main research works of the dissertation are as follows:The platform of the autonomous excavation system is designed according to the specific excavation conditions and functional requirements,and the selection and installation of hardware equipment are completed.A computer control system platform is built for the motion control of the working device and the swing device of the autonomous excavation system,and the sensing and planning program of master computer and control program of the slave computer are developed respectively.An unsupervised point cloud processing-based penetration point detection method is proposed for autonomous excavation.Firstly,the 3D point cloud of the excavation area under the global base coordinate system of the excavator is obtained by the coordinate transformation from the stereo vision technique,and the camera coordinate system to the global base coordinate system of the excavator.Then the point cloud of the excavation area of the trench is extracted based on the unsupervised global graph clustering method.Finally,a global gradient consistency function is constructed to describe the geometric features of the penetration points based on the geometric features of the trenching area contour of the work device plane to achieve the detection of penetration points.The field test shows that the method can effectively realize the extraction of trench excavation area,and its accuracy,recall and F1 score can reach97.69%,93.82% and 95.72%,respectively.Within a 5×5.5 m excavation range,the maximum absolute positioning error of the penetration point is 69.0 mm,and the average relative error is1.36%,which verifies that this method can be used for penetration point detection in autonomous excavation.A method is proposed to measure the position and orientation of the dumping point of autonomous excavation by tracking the marker with a calibrated monocular camera,which can precisely measure the dynamic position parameters of the dumping point and achieve a more accurate measurement of the position of the dumping point during the dynamic excavation operation.The test results and accuracy analysis show that the maximum measurement distance of the system is 11 m,the maximum attitude angle error is 8°,and the maximum position error is 22 mm.Finally,the field test results show that the maximum absolute position error of the dumping is 48 mm,and the average relative positioning error is 0.66%,which verifies the feasibility and effectiveness of dumping point positioning.A method is proposed to generate the motion trajectory of the excavator operation process based on the digging path of the skilled driver.The trajectory of the skilled driver operating the excavation operation is first converted into a topologically equivalent trajectory.Then the excavation trajectory is parameterized by using the spline function,so that the trajectory passes through the path points that are topologically equivalent to the manual excavation trajectory in turn,and the trajectory generation problem is solved iteratively and optimally to obtain the time-jerk optimal trajectory under the kinodynamic feasibility constraint.The field test results show that the time used for the autonomous excavation operation during the 7-shovel excavation is 80.4 s at filled bucket excavation,which is less than that used for the manual excavation of 83.2 s,which verifies the feasibility of this method.A complete vision-based model is built for the trajectory optimization of hierarchical trenching autonomous excavation.Firstly,an excavation model for trenching operations is developed for path planning of the bucket teeth.Then,the position information of the penetration and dumping points is obtained through visual perception techniques,and the intermediate waypoints are obtained by using the topological information of manual excavation operation.The Bézier curve is used to connect these waypoints,and the trajectory is re-time parameterized.A fast,smooth and stable mining trajectory is generated through spatialtemporal optimization under the constraints of a filled bucket and kinodynamic feasibility constraints.Finally,through the comparison of autonomous excavation and skilled manual excavation in field test,the time used for autonomous excavation is 80.8 s,and the time for skilled manual excavation is 84.1 s.The trajectory optimization method can improve the efficiency of excavation operation by 3.9%,and the jerk of autonomous excavation is less than manual excavation,and the trajectory generated by the method can achieve the expected effect in terms of smooth and efficient motion.