Modeling of robot motion as dynamical movement primitives (DMPs) has
become an important framework within robot learning and control. The
ability of DMPs to adapt online with respect to the surroundings, e.g.,
to moving targets, has been used and developed by several researchers.
In this work, a method for handling perturbations during execution of
DMPs on robots was developed. Two-degree-of-freedom control was
introduced in the DMP context, for reference trajectory tracking and
perturbation recovery. Benefits compared to the state of the art were
demonstrated. The functionality of the method was verified in
simulations and in real-world experiments.