Kinesthetic Teaching via Trajectory Optimization with Common Distance Metrics

We propose a method for learning skills by learning kinesthetic skills via trajectory optimization. We recreate demonstrations by optimizing the trajectory to be near the demonstration. We use the trajectory optimizer, TrajOpt to turn distance metrics, the Frechet and Hausdorff distances, into cost functions that penalize a trajectory from deviating from the demonstration. Using a few techniques to assist the optimization process, we are able to reliable recreate provided demonstrations.