Goal Babbling - direct learning of inverse kinematics

For robots with many redundant degrees of freedom or soft robots with
unknown kinematic function it is difficult to learn the inverse kinematics.
Goal babbling refers to a scheme to nevertheless do this online, while behaving and actively generating new information to improve on the fly.
It is motivated from infant motor learning and has successfully been applied to planar snake-like robots up to 50 DOFs as described in Rolf et al. and applied to an soft elephant trunk robot in Steil et al..