Children acquire reusable skills by playing, discovering what they can control before anyone hands them a goal. The same idea powers exploration in embodied AI: intrinsic motivation, a reward the agent gives itself for novelty, surprise, or prediction error. Instead of waiting for an external prize, the agent is drawn to states it has not seen or cannot yet predict, which is exactly how a policy can build a map of its abilities before a task arrives. One competition entry even explored by chasing any frame that changed, downvoting actions that did nothing. Here you build the simplest version, seek the least-visited place, and watch curiosity cover a world that a random walk leaves half-blank.