| The large-scale application of industrial robots in processing and manufacturing greatly improved the production efficiency,but now the complex and changeable production environment requires the collaboration between robots and people to complete a complex production task.The traditional fence structured security protection scheme can not meet the interaction requirements in the human-robot collaboration,so the construction of active security method by means of vision and trajectory planning has become a research hotspot in the field of human-robot collaboration.The human-robot collaboration mode of speed and separation monitoring defined in the design standard of cooperative robot calculates the safety range through speed.As long as there is target intrusion,it will directly ensure safety through deceleration and shutdown,without distinguishing the categories of dangerous targets and affecting efficiency.It is easy to crash due to the rapid intrusion of non-operators,untimely deceleration,or misjudgment and shutdown when operators need close cooperation.In order to extract the danger information in the human-robot collaboration scene comprehensively and accurately,and use the extracted information to adjust the robot motion,so as to ensure the safety and efficiency of human-robot collaboration,this paper studies the scene danger information perception and robot safety control methods.Firstly,in order to solve the problem of unstable and incomplete information perception under different lighting conditions such as dark light and backlight,a multi-information fusion perception network is proposed in this paper.The infrared image and depth image insensitive to light change are used to extract edge features,the RGB image is used to extract category features,and a feature fusion pyramid network structure is designed to fuse and enhance the feature layers of different types and scales,enhance the original RGB image,and improve the detection accuracy of the network in dark light.Then,a dynamic speed and separation monitoring algorithm based on semantic information and an S-PRM algorithm based on semantic feasible space are proposed to integrate the obtained semantic information into the security regulation of human-robot collaboration.The classification,location,quantity and other information of danger information obtained by neural network are transformed into risk factors,so as to avoid collision with dangerous targets by deceleration,path planning and shutdown according to the danger degree of the scene,and realize multiple protection for the robot on the premise of reducing efficiency loss.Finally,a human-robot collaboration system integrating speed regulation and trajectory planning is built.At a long distance,the robot is kept away from danger as far as possible through the selection of safe space.In case of medium and short distance,the danger can be avoided by deceleration and shutdown.The multi-level protection of the robot is realized by deceleration,path planning and shutdown.In this paper,the actual effect of the algorithm is tested under different lighting conditions.The experimental results show that the security system proposed in this paper can effectively improve the security of man-machine cooperation. |
