Research On Probabilistic Robotic Hand-eye Calibration Algorithms

Robot hand-eye calibration algorithm is the basis for the robot perception of the external world,and it is also an important prerequisite for other robotic algorithms.When the robot senses an external object through sensors,the hand-eye calibration algorithm can perform the pose of the external object relative to the robot coordinate system."Hand" usually refers to the end effector of the robot,and "eye" refers to the sensor,such as a camera,etc.Many researches on perception algorithms need to obtain accurate hand-eye relationship as the basis.The probabilistic hand-eye calibration algorithm is the main direction of the current theoretical development.It has the advantages of fast calibration speed,large amount of data and high accuracy,so it can be applied to more occasions.Therefore,researching on probabilistic robot hand-eye calibration algorithm is of great significance in theoretical models and practical applications.This paper divides the factors of accuracy in probabilistic hand-eye calibration algorithm into two parts: theory and calibration data set.At the level of algorithm theory,a probabilistic hand-eye calibration algorithm under the definition of a new mean is proposed in this paper.The author analyzes the mathematical derivation and theoretical assumptions of the current “Batch” series of calibration algorithms,and summarizes their existing problems and limitations.According to the assumptions,the current mean of the translation part and mathematical model should be modified to improve the accuracy of the translation part of the algorithm.Moreover,the hypothetical limits of Normal Distribution should be removed.Simulation experiments prove that the revised definition of mean is effective.At the level of calibration data sets,there are misalignment problems like timeshift and data errors in the process of collecting large amounts of data.This article first classifies and summarizes the data misalignment problems caused by different sensor frequencies and the asynchronous of the sensors.According to the time stamp acquisition situation,four solutions are given,and the theory of the interpolation algorithm and cross-correlation method are given in detail.Error analysis is performed on the collected data and a data screening method based on the idea of random sampling is proposed.Such algorithm filters out the calibration data which yields large errors by defining the degree of preference in each data pair.The combination of the screening algorithm and the new definition of the mean together constitute the hand-eye calibration algorithm proposed in this paper.Finally,an endoscopic surgical robot system and a 3D camera industrial robot system are used to verify the proposed hand-eye calibration algorithm.Two methods for data collection are given respectively.The error comparison between the proposed algorithm and the aforementioned Batch series methods and traditional calibration methods in terms of translation and rotation is performed.The results show that our algorithm which improves on the Batch 1 method has good performance in the accuracy of the translation part.The proposed algorithm has a certain versatility,which can be better used in calibration occasions.