Abstract:
Industrial manipulators are capable of completing a wide range of tasks. One of
these tasks is object manipulation. There exist two ways to perform such tasks. The
first is to have a fully defined problem with known kinematic and dynamic models
of the manipulator and the objects. This approach has been shown to be successful
in a huge number of works. However, it has some limitations connected to cases
when we do not have complete knowledge about the robot, object, or the task itself.
Thus, another approach is to use machine learning in order to learn to perform a task
without knowing its intrinsic parameters. In this work, we research both approaches
and show how we can use them in order to perform some dexterous manipulation
tasks. The former is illustrated in the example of the human-robot object handover.
The latter is shown in a variety of experiments where a robot learns to perform a
specific task.