Adaptive Autonomous Navigation of Mobile Robots in Unknown Environments

Autonomous navigation of a mobile robot is a challenging task. Much work has been done in indoor navigation in the last decade. Fewer results have been obtained in outdoor robotics. Since the early 90's, the Global Positioning System (GPS) has been the main navigation system for ships and aircrafts. In open fields, satellite navigation gives absolute position accuracy. The absolute heading information is also obtained by satellite navigation when the mobile robot is in motion. However, the use of GPS satellite navigation is mainly restricted to open areas where at least three satellites can be seen. For example, mobile robots working in underground or deep open mines cannot use satellite navigation at all, and in forest or city areas, there are serious limitations to its use. Laser range finder technology has evolved remarkably over the last decade, and offers a fast and accurate method for environment modeling. Furthermore, it can be used to define robot position and heading relative to the environment. It is obvious that the use of several alternative sensors according to the environment will make the navigation system more flexible. Laser range finder technology is particularly suitable for indoors or feature rich outdoor environments. The goal of this thesis is to develop a multi sensor navigation system for unknown outdoor environments, and to verify the system with a service robot. Navigation should be possible in unstructured outdoors as well as indoor environments. The system should use all available sensor information and emphasize those that best suit the particular environment. The sensors considered in this thesis include a scanning laser range finder, a GPS receiver, and a heading gyro. The main contribution of the thesis is a flexible navigation system developed and tested with a service robot performing versatile tasks in an outdoor environment. The used range matching method is novel and has not been verified earlier in outdoor environments. No unique solution can be guaranteed in the developed map matching algorithm, although it seems to work well in the practical tests. Position and heading errors grow without bound in successive map matchings, which could be referred to as laser odometry. Therefore, the position and heading have been corrected by means of global matching when the robot returns to a place it has previously visited. Alternatively, structured landmarks have been used for position and heading correction. In field tests, tree trunks and walls have been used as structured …