Before any task or reward, children play, and through play they discover what they can control and build a repertoire of skills. It is not idle; it is a self-directed curriculum that samples goals which are novel, meaningful, and just beyond current ability. Modern embodied AI formalizes this as intrinsic motivation and learning progress: an agent that practices near the boundary of its competence expands that boundary fastest, while one that samples goals at random mostly attempts the impossible and learns little. This is why play is a big part of existence, for a child or a robot: it is how a body finds out what it is capable of. You implement the two strategies and watch play win.