next up previous contents
Next: A Biologically-Motivated Gaze Up: Active Perception Previous: Gaze Planning: A

Mobile Robot Navigation and Landmark Recognition Using QUADRIS

The main tasks of an autonomous mobile robot are path planning and navigation. A map of the environment is necessary for these tasks, and there are several ways one can be produced. It can either be explicitly predefined to the robot, or in an unknown environment, be built dynamically. We are interested in entering uncharted areas, so it is imperative that the robot be able to build an accurate map of its immediate surroundings in order to plan its next step. This research deals with a new sensor platform, QUADRIS, which is ideal for these tasks as well as for object recognition. QUADRIS is a range sensing system comprised of two independently controlled laser rangefinders, which have two degrees of freedom each, a pan and a tilt movement. The QUADRIS platform is mounted on a mobile robot, which is sent into the environment to be scanned to retrieve range data, which can then be used for various purposes. Having this flexibility facilitates covering as much of the robot's surroundings as possible during map building. The environment we are working with is an office space. The tasks are mobile robot path planning, navigation, and landmark recognition of structures and objects. QUADRIS provides the range data for navigating through the hallways of an office environment. A second task is the recognition of landmarks that are generally found in an office space. These landmarks can be categorized into the following groups: (i) structural objects, such as doors and walls (ii) movable objects, which basically constitutes furniture such as desks and chairs (iii) parametric geons, such as blocks and cylinders, which are specific shapes placed in the environment. The recognition process entails constructing a ``viewing sphere'' object model, using relevant features extracted from 3-D models of the objects listed above. A viewing sphere is an imaginary sphere centered about the geometrical model of an object, and represents all the different possible views of that object. The surface of the sphere is divided into regions from which the object appears to share the same features. The distinct qualitative structure of the view from each of these regions is called an ``aspect''. Aspects may be defined in any number of ways. To define our aspects we use features obtainable from QUADRIS, namely the presence of geometric planes on the object, its height and its material construction. The recognition process is then a comparison of a view of an unknown object with the data stored in the viewing sphere model. Practical applications for this system include different security needs, such as office and warehouse surveillance.

D. Bui, M.D. Levine



next up previous contents
Next: A Biologically-Motivated Gaze Up: Active Perception Previous: Gaze Planning: A



Thierry Baron
Mon Nov 13 10:43:02 EST 1995