Noise is the default, not the exception
New robotics students often assume sensors return the truth and the software just has to read it off. In reality, every sensor reading has noise: a LiDAR beam that grazes a table edge might report a distance that's off by centimeters, a camera image gets grainy in low light, and an IMU accumulates tiny drift errors every second it operates. None of this means the sensors are broken, it's simply the physical reality of measuring the world with imperfect instruments.
Good robotics engineering assumes noisy data from the start rather than treating it as a bug to eliminate. That's why techniques like sensor fusion, filtering, and probabilistic reasoning exist. A robot that only worked with perfectly clean data would fail the instant it met the real world, which is precisely why so much of robotics software is built around handling uncertainty gracefully instead of assuming it away.