Sensor fusion for localizing a mobile robot outside of buildings

In this paper we show that it is possible to extend a mobile indoor service robot, making it capable of performing autonomous transportation talks in outdoor environments. While the higher-level software (like planners, schedulers and collision avoidance) is no different than that used in indoor vehicles and therefore both well known and reliable, the sensor fusion layer is the challenging part due to the extreme ambiguity of sensor data outside most buildings. Therefore we focus an the self-localization aspect of the system, showing that it is possible to predict the robot's position with such low uncertainty that a typical transportation task in an industrial outdoor environment can be performed quickly, safely and robustly. We rely only an low cost off the-shelf sensors (optical encoder, fiber-optical gyroscope and a laser range finder), without adjusting the robot's environment (e.g. by way of addition of artificial landmarks), making the system affordable and easy to maintain. We prove the usability of our approach through extensive tests with a real robot around our institute campus. Test drives totalling over 20km show that it is possible to find and traverse a target door in over 99% of all cases.