An Investigation of Different Modeling Techniques for Autonomous Robot Navigation

This research aims to give recommendations towards modeling the navigation control architectures for an autonomous rover designed for an unstructured, outdoors environment. These recommendations are equally applicable to other autonomous vehicles, such as aircraft or underwater vehicles. Many successful architectures for this application have been developed, but there is no common terminology for the discussion of robotics architectures and their properties in general. This paper suggests the use of terms borrowed from administrative theory to facilitate interdisciplinary dialog about the tradeoffs of various kinds of models for robotics and similar systems. Past approaches to modeling autonomous robot navigation architectures have broken the architecture up into layers or levels. The upper levels or layers make high-level decisions about how the robot is going to accomplish a task, and the lower levels or layers make low-level decisions. This is analogous to a CEO of a corporation telling the managers how he wants the corporation to work towards its goal. The managers each oversee a part of the corporation. The workers are told what to do, but still make low-level decisions such as how hard to twist a screw, what tool to use to remove a rivet, or to do something other than what they were told in the interest of safety. Traditionally, there have been two or three layers for robot architectures, and every module developed fits into one of these layers. Every branch of the hierarchy has one module in each of the layers. The reasons given for breaking the architecture up into two or three layers vary fiom implementation to implementation. This paper aims to take a more