engleski

On-line Estimation of F/T Sensor Offset for Arbitrary Orientation of Robot Tool by Evaluating Two Machine Learning Algorithms

The offset in the output of a force/torque (F/T) sensor embedded in the robot's wrist is usually handled by deducting the measured force value when the robot tool encounters the surface and achieves the desired orientation. The problem becomes more complicated when the tool's orientation changes continuously or the tool is changed during the robot's work. This paper explored the possibility of using two well- affirmed machine learning algorithms (k- nearest neighbors and polynomial regression) for on-line estimation of F/T sensor offset. Both machine learning algorithms were implemented and evaluated on the state-of-the- art 6-DOF robot arm Schunk LWA4P equipped with the OptoForce HEX-70-CE-2000N F/T sensor. Evaluation of the two on-line offset estimation methods was made during a force-controlled writing on the office board with different robot tool orientations. It was found that the frequency of generating the offset estimate is much higher with the polynomial regression algorithm.

estimation theory , force control , force sensors , learning (artificial intelligence) , manipulators , nearest neighbour methods , polynomials , position control , regression analysis

nije evidentirano