Comparing feature-based and GAN-based methods for learning from demonstrations, the key to success lies less in the method itself and more in the structure and suitability of the motion representation for the task.
