Modern industrial robots are increasing rapidly towards collaborating and physically interacting with people on complex tasks, and away from working in isolated cages that are separated from people. Collaborative robots are usually developed to perform variable stiffness control, teaching by touch, collision detection, and so on. Torque control with accurate gravity compensation becomes necessary.
When the method of torque-based impedance force-control is used to achieve teaching by touch, the gravity compensation is added to the joint torque loop directly. As a result, the gravity model should be very accurate, otherwise the robot will not stay still even if no any artificial external force is applied. Unfortunately, it is very difficult to model and identify the robot dynamics such accurately in the whole working space because of the presence of the robot cables, joint and link flexibility, cables or tube of end-effector, the end-effector itself and so on. There are always some regions where the robot will drift due to the error of gravity compensation.
This paper is motivated by the gravity compensation problem of collaborative robots equipped with joint torque sensors, and attempts to propose an accurate on-line correction strategy for gravity compensation especially aiming at the application of teaching by touch. The developed method has been tested on a 7-DOF collaborative robot and shown good performance.