Research On Environmental Perception And Path Planning Algorithms Of Collaborative Robots Based On 3D Vision

In recent years,with the increasing level of intelligence in industrial production lines,on the one hand,traditional industrial robots have shown high efficiency and stability in practical applications such as palletizing,automatic assembly,and welding;on the other hand,the deployment of robot production lines The cost is high,the cycle is long,the adaptability to small batches,customized products is weak,the safety is not strong,and the isolation fence needs to be set,which reflects the shortcomings of traditional industrial robots.As its new branch,collaborative robots have the advantages of safety,low cost and easy operation through man-machine collaboration strategies,and become the development direction of intelligent production lines.However,in the face of the diversification and uncertainty of the work scene,the collaborative robot adopts offline teaching or the construction of the scene model is not universal,and there are often some dynamic or unknowing obstacles in the collaborative space,such as operation People or other objects that fail to make evasive measures may cause the tragedy of plane destruction and death.Therefore,the environment perception and path planning of collaborative robots in unknown environments have become practical problems that need to be solved urgently.Based on this background,this paper proposes a three-dimensional vision-based collaborative robot environment perception and path planning method.The research uses three-dimensional sensors to acquire the spatial pose of the target workpiece and build the discretization model of the work scene.Based on the improved Rapidly-exploring Random Trees algorithm,the collaborative robot can grasp and place without collision in an unknown obstacle environment.The main work is as follows:(1)According to the actual needs of the industrial production line,construct the overall system plan,including system frame design,system software and hardware plan and selection and algorithm flow design;at the same time,for the subsequent positioning and coordinate conversion of the reconstruction module,the vision system calibration is explained;(2)Achieve target recognition and positioning based on a three-dimensional sensor.First,perform preprocessing operations on the collected images,and design a recognition and positioning system: To obtain the precise pose of the target,we use the template matching algorithm to deal with small and simple objects based on SURF(Speeded-Up Robust Features)feature points,and use the positioning method to deal with large and complex objects based on Ar Uco QR code;through experimental analysis,the recognition and positioning errors of t he two cases are maintained within the range of 3mm and 8mm respectively;(3)Use the vision system to build a discretized model of the work scene.Through the three-dimensional sensor at the end of the robotic arm,the depth point cloud of the work scene is obtained,and the pose of the end of the robotic arm is merged,and the global environmental point cloud is obtained through preprocessing,matching and calibration,and model optimization;then based on the octree data structure,a discretized model,Mapped to the subsequent planning space;and analyzed the modeling accuracy,and the size and positioning accuracy were maintained within the range of 2mm and 8mm respectively;(4)Improve the strategy of the planning algorithm to improve the obstacle avoidance performance of the manipulator.On the basis of the shortcomings of traditional RRT(Rapidly exploring Random Trees)and derivative algorithms,the Cauchy distribution sampling,target gravity and node rejection strategies are introduced,and experimental analysis is carried out based on the six-dimensional space of the robotic arm under matlab and ROS(Robot Operating System).The statistical planning success rate is 96.5 %.(5)Design the controller framework,software system and application experiment.Under the HIROP(Huashu intelligent robot operating platform),based on the joint control system of Move It,Gazebo and OMPL(Open Motion Planning Library),the complete obstacle avoidance planning experiment of the real manipulator is carried out.The average time for identification,modeling and planning is maintained within 5s,and the obstacle avoidance success rate exceeds 98%,indicating that this The algorithm is effective and reliable.