| Live-line maintenance is safeguard for power supply of industrial production and residents’ living.It is urgent to improve the automation level of power maintenance system of our country.Development of autonomous live-line maintenance robot is of great significance to construct efficient,safe and reliable power maintenance system.In this paper,based on the research and development project of autonomous live-line maintenance robot,and on the basis of the overall design of the robot system,key technologies such as automatic hand-eye calibration,the posture perception and manipulation of the flexible cable,the peg-in-hole assembly under visual occlusion,and the robot task-level planning are studied.The main contents and innovations are as follows:Aiming at the problems of complex terrain and unstructured environment in live-line maintenance,one kind of robot system consisting of self-propelled aerial working vehicle,ground control room and aerial robot platform is constructed.In this paper,the control system,communication system,mechanical system,sensor system,power system,insulation protection system are designed for the difficulties and mission requirements.According to the structure of the robot system,the task flow of live-line maintenance is further refined,which lays a foundation for the subsequent study of robot control strategy.An automatic hand-eye calibration method based on variant particle swarm optimization is proposed to meet the requirement of rapid hand-eye calibration for complex live-line maintenance.Based on the camera imaging principle and calibration method,the depth data correction is designed,which improves the accuracy of 3D measurement.As the routine points of manipulator for hand-eye calibration rely on manual work,which has the low efficiency,the hand-eye parameters can be estimated real-time in the process of manipulator movement based on variant particle swarm algorithm,and the joint angle of manipulator can be compensated based on energy optimization principle,which ensures that the calibration plate is always in the visual field of the depth camera during the calibration process.Experimental results verify the effectiveness and accuracy of the proposed algorithm.On account of the most typical requirement of flexible cable grasping in live-line maintenance,the cable pose perception based on symmetric edge feature and grasping planning based on operation surface are proposed.Aiming at the problem of the uncertain shape of spatial flexible cable and the difficulty of establishing its mechanical model,Symmetric Edge Features(SEF)are designed based on cable projection principle,and the edge detection algorithm is designed based on semantic segmentation to reduce the computation cost.For images containing cables in the complex background,edge extraction,feature filtering,center point sorting and center line fitting are carried out for calculating its precise 3D poses.In view of the fact that the slender cable will increase the probability of collision in the grasping process,the cable grasping operation surface is designed,and the cable grasping point is selected according to the force directional manipulability.Then a two-phase planning algorithm based on the BGRRT Guided by Motion Directional Manipulability(MBG-RRT)is proposed.Simulation and experimental results verify the efficiency and accuracy of the proposed algorithm.Aiming at the assembly problem of crimp terminal under the condition of visual occlusion and uncertain mechanical model,an assembly strategy based on Generative Mapping and Search Network(GMSN)is proposed.In this paper,the assembly process is divided into two phases,rough alignment and precise hole-searching.In the phase of rough alignment,the target is not blocked.The deep convolutional neural network is used to detect the target position and realize the visual servo.In the phase of precise hole-searching,when the target is partially blocked,a two-dimensional hole-searching model is constructed based on visual and force information,and a new self-supervising network-GMSN is designed to achieve accurate holesearching.Simulation and experimental results show that the live-line maintenance robot can accomplish the crimp terminal assembly task safely and efficiently under the proposed strategy.In order to realize the autonomous manipulation in unstructured environment,the tasklevel planning of robot system is further explored.The scenario state and action primitives of the robot are abstracted into nodes and edges based on the knowledge graph,and the liveline maintenance task map is constructed according to the physical constraints and task flow constraints.A scenario state estimation method based on Dynamic Bayesian Network(DBN)is proposed by the connection between nodes of task map and real-time detection of key targets in the scenario by visual system.A value iteration network based on graph is designed combinating the idea of reinforcement learning.This network has the characteristics of high efficiency,stability and strong path reuse.Experiments in laboratory and outdoor live-line maintenance environment verify that the task-level planning scheme enables the robot system to sense the scenario state with high accuracy and make the optimal decision. |
PDF Full Text Request