Dynamic Movement Primitives (DMP) for goal-directed motion generation

DMPs implement the idea to utilize second order dynamical systems, specifically spring-damper dynamics, to generate generic converging motion towards a particular goal value in operational or joint space.
They receive ample attention in imitation learning, because this generic motion can be shaped by adding an additional force term in the equation, which can easily be learned from human demonstration data. The standard approach is described in Schaal et al. and it has been applied to learning many tasks including table tennis, ball-in-cup, or shooting balls.