Research On Genetic Algorithm Apply To Identify The Robot's Kinematics Parameters

With the development of modern industry technology, higher pose (position and orientation) accuracy of the robot end effector is required. However, the traditional industrial robots generally work through being instructed in the off-line way since they only finish simple tasks. Correspondingly, most people only care for the robot's repeatability. Good repeatability only shows enough compactness of the robot'part and enough resolvability for joint encoders, while it can not necessarily represent the right transformation relation between the robot's joint angle and its end effector's pose. If the robot's higher repeatability is changed into higher end effector's pose, we need identify the robot's kinematics parameters, namely robot calibration.Firstly, aiming to independence of the robot's kinematics parameters, the paper has introduced such the three concepts as completeness, singularity and equivalence. Among them, the robot's completeness is emphatically discussed. According to whether the joint quiver or not, the robot's kinematics models can be divided into the conventional models and the generalized models. On the basis of geometrical analysis and error space analysis, the calculation formulas of the previous two models are derived.Secondly, because D-H modeling method will cause the kinematic parameters of the consecutive parallel joints relative, the improved kinematic transformation matrix between the two links is presented. Subsequently the pose error model between the consecutive links is derived. Then, the pose error model of the robot's end effector is built. According to complexity, the model is simplified into the first-order pose error model and the second-order pose error model. Based on these, the end effector's pose error distribution is discussed and the first-order error envelops and the second-order error envelops are calculated.Thirdly, the paper puts forward an improved genetic algorithm based on Crowding-out mechanism. The algorithm delete the similar individual based on similarity between the group's individual so as to maintain the population diversity and refrain the algorithms from premature convergence. considering the influence of the measurement noises, in the light of the above pose error model and the different simulation types categorized based on the measurement position number and the repeated measurement times, we carry out the kinematic parameter-identification simulation respectively using the least square and the genetic algorithm(GA), which shows that the robot end effector's pose accuracy is improved obviously and proves the validity of these algorithms. Simultaneously, the merit and shortcoming of each algorithm are analyzed according to these simulation results. Attentively,it is found that the GA intensively adapts to the kimematic model and is not sensitive to the relative kinematic parameters because of its randomicity and uncontinuity.