Gaussian distribution (in estimation)

A Gaussian distribution is the famous bell-shaped curve: high and humped in the middle, tailing off smoothly toward zero on both sides. In estimation it is the favorite shape for describing a robot's belief, because it captures the two things that matter most about a guess with just two numbers. The first number is the mean — the location of the peak, the robot's single best estimate, its "I think it's right about here." The second is the spread, which says how wide the hump is — narrow and tall for a confident robot, broad and low for an unsure one.

That spread is captured by the variance (and, for a guess about several quantities at once, the covariance matrix); the everyday "plus-or-minus" width you picture is its square root, the standard deviation. The beauty is the economy: instead of tracking thousands of separate possibilities, the robot stores just a center and a width and trusts the bell curve to fill in everything between. Real-world noise also tends to pile up into roughly this shape on its own — many small, independent errors jostling together average out into a bell — so the Gaussian is not just convenient, it is often a genuinely faithful picture of how uncertainty actually behaves.

There is one more gift that makes engineers love it. When you start with a Gaussian belief and push it through the kind of straight-line (linear) math that simple motion and sensors involve, the result is still a Gaussian — the bell may shift and stretch, but it never turns into some lumpy, two-humped monster. This closure under linear operations is exactly why the Kalman filter can run so fast and cleanly: it only ever has to update a mean and a covariance, never an unwieldy cloud of points.