A known map doesn't mean a known position
It's tempting to think that once a robot has a good map, knowing where it is on that map should be easy. It isn't. A map tells you the shape of the world, but it says nothing about where within that world the robot currently sits, and every source of information a robot has for figuring that out is imperfect. Wheel odometry, tracking position by counting wheel rotations, drifts over time because tiny errors in wheel slip, uneven floors, and rounding accumulate with every turn, so a robot that trusted odometry alone would slowly diverge from reality even while sitting on a perfectly good map. GPS, when available at all, carries several meters of error and vanishes entirely indoors or between tall buildings.
Even instantaneous sensor readings, a single laser scan or camera frame, are individually ambiguous. A hallway that looks identical for twenty meters gives a laser scanner no way to distinguish position 3 from position 13 within that stretch. Taken alone, almost no single measurement pins down a unique position. The honest conclusion is that a robot can never know its exact position with certainty, it can only maintain a probability distribution over where it likely is, and get better at narrowing that distribution over time.