Adapting periodic motion primitives to external feedback: Modulating and changing the motion

Abstract

Learning and execution of trajectories using dynamic movement primitives (DMPs) incorporates properties, which make them widely accepted and used in synthesizing robotic motions. The properties include fast, robust and numerically undemanding learning on one side, and indirect dependence on time, response to perturbation and possibility to modulate during the execution. Modulation properties include both spatial and temporal changes to either discrete or periodic motions. In this paper we evaluate the means of adapting periodic motions using either force or position feedback in order to permanently modify the motion, i. e. learn a new trajectory in order to comply with the conditions of the external environment. We evaluate three different approaches: a modulation approach using repetitive control; and two learning approaches of changing the motion. Simulation results have shown that all three approaches can be used with minor differences amongst them. Tests on a 7DOF KUKA LWR robot have shown that the approaches can be used in the real-world.