Robust Small Robot Localization From Highly Uncertain Sensors

Localization is arguably the most important goal for a robot to solve-without knowledge of its place in the world, a robot cannot do useful work. The current best practice for providing accurate localization given uncertain sensors involves the use of sensor fusion, or combining sensor data in order to derive a better pose estimation for the robot. Small robots add another problem that needs to be solved-limited power, computational ability, and weight/space constrain both the sensors available and the filters that can be used. This paper provides a detailed example in simulation of four filter types: the extended Kalman filter, the Fuzzy EKF, the sigma-point KF (SPKF), and the double fuzzy SPKF, while discussing the strengths and weaknesses of all current state of the art sensor methods. While the field is relatively mature, there has been little to no comparative analysis of different filters performed-what roles do they best serve, how to select the “best” filter, and what tradeoffs must be made for each type. This paper analyzes the current state of the art filters of all categories and determines their applicability to the small robot problem.