Probabilistic Approach to Modeling and Parameter Learning of Indirect Drive Robots From Incomplete Data

Abstract

This paper deals with the problem of modeling and identification for industrial robots with indirect drive mechanisms (e.g., gear transmissions), where the motor-side behavior may deviate from the load-side (i.e., end effector) behavior due to the robot joint dynamics. In this case, the motor-side sensors alone are not sufficient for directly obtaining the load-side behavior. We call such problems incomplete data problems since some key information (i.e., the load-side behaviors) is missing during the identification process. This paper, thus, presents a procedure for identifying robot models even with limited load-side information. It begins by describing the relationship between the missing information and the available measurements using a Gaussian dynamic model. Then, the expectation-maximization algorithm is applied to handle the parameter estimation problem in the presence of missing information. Various relaxation techniques are utilized to alleviate the computational complexity of the algorithm. The effectiveness of the proposed method is demonstrated on three real-world problems in industrial automation, namely the gain tuning of the Kalman filter, the friction identification, and the self-calibration of load-side inertial sensors.