Study Of Robot Peg-in-hole Assembly Based On Human Demonstration

With the development of robot automation technology,more and more work in factories can be replaced by robots.Howerver it usually leads to failure of assembly with the traditional robot programming method,which is less compliance.Compared with the robot,human beings can complete the assembly task flexibly and quickly.The introduction of human experience in teaching robots can solve the above-mentioned problem of assembly.In this paper,we intends to study the learning control problems in all stages of the robot peg-in-hole process,and make the robot has the humanoid assembly capabilities.According to the characteristics of peg-in-hole assembly,the assembly movement is divided into two stages: approaching phase and pose adjustment phase.A control model based on dynamic movement primitive algorithm is established in the approaching phase.Considering the obstacles may exist in the approaching phase,the artificial potential field function is added to the dynamic movement primitive model for obstacle avoidance,which ensures the robot to move fast and safely in the space.The pose adjustment is the most critical stage in the assembly process.Firstly we analysis contact force to determine the important control variables in assembly movement.Aiming at the asynchrony of teaching data,the dynamic time warping algorithm is used to unify the data in time series.Based on the Gauss mixture model,the nonlinear mapping relationship between the control variables and the state variables of the shaft is established,and the parameters of the model are solved by the EM algorithm.In order to verify the effectiveness of the teaching lear ning algorithm,the peg-in-hole assembly experiment was carried out on the Barrett WAM robot.Different controllers were designed for the two stages of assembly.In order to increase the flexibility in the assembly process,the impedance controller was added between the learning controller and the robot to improve the success rate of assembly.Finally,the experiment shows that the controller based on the teaching learning not only improves the efficiency of robot assembly,but also makes the robot adaptive to complex environment.