Research On Human Robot Collaborative Assembly Technology Based On Human Skeleton Motion Recognition

Robot assembly technology has been widely applied in workshop due to its efficiency and repeatability.However,in some complex assembly processes or flexible operations,manual assembly cannot be completely replaced because of its flexibility and ability to handle the unexpected.Human robot collaboration allows people and robots to work independently in the same space,and can give full play to their advantages.It is an important research direction of intelligent assembly.In the human robot collaboration task,the robot needs to understand the assembly behavior of workers and provide auxiliary operation.How to accurately and effectively identify assembly behavior is a difficulty in this task.On the other hand,the current human robot collaboration research mainly focuses on triggering a robot command through recognized behavior or gesture according to a fixed assembly sequence,without considering that workers can subjectively adjust the sequence.This paper studies the above problems,and the main research includes:(1)A method of assembly behavior recognition based on fusion features was proposed: human skeletons were obtained based on lightweight Openpose and 1€filter.The spatial feature and temporal feature were extracted respectively through linear layer and self-attention model.A pre-trained fully convolutional network was used for the feature of the workpiece.Features were fused with the self-attention model and the model was trained with soft label labeling method.(2)A dynamic human robot collaborative assembly method based on assembly andor-graph is proposed.The positions of workers and parts in the collaborative scene are identified and their coordinates are unified to the base coordinates of the robot.Based on the traditional assembly and-or-graph,the assembly behavior nodes are integrated and the part tables and behavior tables are constructed to store the real-time assembly information.A dynamic traceability algorithm based on parts behavior assembly and-or-graph is proposed to provide auxiliary operation for workers according to the recognized behavior.(3)Based on the assembly case of turbine reducer,the parts-behavior assembly andor-graph are constructed,and the assembly action dataset is established.Contrast experiment was conducted to verify the performance of our method.After the training model is obtained,human robot collaborative assembly experiment is carried out on UR5 robot collaboration platform to verify the effectiveness of the proposed dynamic human robot collaboration method...