These models typically predict actions directly, there is no motion planning going on here.