Uncertainty is a catch-all term that refers to any and all of the different ways in which the sensors, models, and actuators of a robot do not behave ideally. Sensors rarely give exact and noise-free measurements, environmental models rarely capture all information that matters, and a wheel that does not slip has never been built. As is the case for so many other human capabilities, we rarely notice our own competence dealing with noisy, missing, and conflicting information. This chapter covers some elementary concepts of random processes and then presents classical techniques that can be used to make robots more competent in our uncertain world.

Uncertainty models can be applied to many forms of error, whether the errors are random or systematic, temporal or spatial. Every measurement has some amount of random noise impressed on it as a matter of basic thermodynamics. Modelling that noise can lead to systems of considerably higher performance. Such performance improvements may lead to improved safety, availability, and robustness.