Figure 1.

Planning Movement Primitives are endowed with an intrinsic planning system, which performs inference in a learned graphical model. States are denoted by x_t, controls by u_t, and the time horizon is fixed to T time-steps. In this example the graphical model is used to infer the movement by conditioning on two abstract goals g^[1] and g^[2], which are specified in the learned intrinsic cost function L(τ; θ).