Research On Human-robot Cooperative Task Planning Method Based On Action Recognition

Human-robot collaborative manufacturing refers to the process that human and robot share the same workspace and complete manufacturing tasks.It combines the advantages of robot and human and improves the production efficiency.Modern manufacturing operation has the characteristics of multiple varieties and small batch.human-robot cooperation can not only make use of human flexibility and decision-making ability,but also give play to the advantages of robot anti fatigue,high load and continuous and stable operation efficiency.In the human-robot cooperation scenario,human beings will accumulate fatigue and affect the cooperation efficiency under long-time repeated work.In the traditional task arrangement process,human factors engineering elements cannot be fully considered and the task arrangement can be changed dynamically.At present,there are some defects in the method of human-robot cooperative task planning,such as insufficient human factors,insufficient system dynamics,weak real-time of task planning and so on.Therefore,it is of great significance to study human-robot cooperative task planning and dynamically adjust the assembly task rhythm to stabilize the human-robot assembly rhythm.To solve the above problems,this paper focuses on the dynamic planning of human-robot cooperative assembly task.The main research work is as follows:(1)A human-robot collaborative assembly model based on personnel fatigue is constructed.This paper establishes the part assembly information model and human fatigue model for the assembly process in the human-robot collaborative scene.On this basis,this paper combined with the different advantages of human and robot in the collaborative assembly scene,a human-robot collaborative assembly model based on human fatigue is proposed.(2)Based on the human-robot cooperative assembly model,this paper realizes the dynamic monitoring of human fatigue and the real-time identification of process state in the cooperative scene.The neural network algorithm mediapipe and Yolo V4 algorithm are applied to build a dynamic evaluation model for human-robot cooperative tasks.In this model,the human fatigue detection algorithm based on mediapipe is adopted to realize the recognition of human fatigue;The human hand motion recognition in the assembly process is realized by particle swarm optimization support vector machine PSO-SVM algorithm and the three-dimensional coordinate information of 21 points of the hand;An assembly tool recognition algorithm based on Yolo V4 algorithm in collaborative assembly scene is proposed.Combined with the hand action recognition results,the assembly process is detected;The experimental results show that the comprehensive recognition speed of the model reaches 24 FPS and the comprehensive recognition success rate reaches 91.28%.(3)A dynamic evaluation model for human-robot cooperation task is proposed,and a multi-objective optimization problem of "cost time difficulty" is proposed.Based on the original version of artificial bee colony,this paper improves the feasible decoding,population initialization strategy,elite bee colony generation strategy,following bee generation strategy and local search strategy.Compared with the other three commonly algorithms,this algorithm has better convergence performance and solution stability,and meets the requirements of man-machine task planning in man-machine cooperation scenarios.(4)A human-robot cooperation system for human-robot cooperative assembly is built.Using the hardware equipment such as cooperative robot and camera,a human-robot cooperation system for human-robot task planning in the scene of cooperative assembly is designed.The experimental results show that the proposed algorithm slows down the accumulation effect of assembly on human fatigue through a more reasonable task arrangement,transforms repeated fatigue into intermittent fatigue,and improves the assembly performance by 23%.