Learning Floppy Robot Control

We present a visual servoing method based on a learned mapping between feature space and joint space. We show real world result showing that this control scheme can be used to control a low-cost robotic arm without joint-feedback. We have, for completeness, included a heuristic gripper recogni- tion algorithm. This is done to be able to present and evaluate a complete method that simultane- ously learns the appearance and control of a low cost robotic arm. Our real world evaluation of the accuracy of our presented method shows promising results.