Learning control laws with DS

This lecture covers the concept of learning control laws with Dynamical Systems (DS) for robots. It explains why learned dynamics with standard machine learning algorithms can be unstable, introduces Stable Estimator of Dynamical Systems (SEDS) and Linear Parameter Varying Dynamical Systems. The lecture demonstrates modeling a robot as a point mass moving according to a time-invariant autonomous dynamical system, discusses training data, and the use of Support Vector Regression and Gaussian Mixture Regression. It also touches on fitting using Neural Networks and concludes by emphasizing that learning a control law as a regression problem involves regressing on the velocity given the state.