Learning Attractor Landscapes for Learning Motor Primitives