The focus of this thesis is on autonomous robot navigation in real-world, unstructured environments. By this we mean the ability to move purposefully and without human intervention in environments that have not been specifically engineered for the robot. It is generally recognized that three of the major sources of difficulty in this task are: (1) the pervasive presence of uncertainty, e.g., in sensing and prior information; (2) the need to coordinate activities aimed at different goals, e.g., moving to a given location while avoiding unexpected obstacles; and (3) the need to integrate processes at different levels of abstractions, e.g., strategic planning and low-level control.

Fuzzy logic, a mathematical formalism based on the theory of fuzzy sets, provides tools that are of potential interest here. First, fuzzy logic is the basis of fuzzy control, which is used in an increasing number of applications characterized by large uncertainty. Second, fuzzy logic offers a wide range of aggregation operators, that can be used to trade off different goals. Finally, the intrinsic ability of fuzzy logic to integrate numerical ("fuzzy") and symbolic ("logic") computation suggests its use as a formalism to integrate numeric control and symbolic planning. While many applications of fuzzy control have appeared in the autonomous robotics literature, these other uses of fuzzy logic have received little attention to this date.

This thesis explores some possible uses of fuzzy logic for autonomous robot navigation. To do so, we develop and validate solutions based on fuzzy logic to some instances of the three problems above; we do so in a behavior-based framework. This thesis makes the following contributions:

1. A practical solution to the problem of the heuristic design and implementation of simple navigation behaviors, based on the techniques of fuzzy control;
2. A new solution to the problem of behavior coordination, based on the use of fuzzy arbitration rules and fuzzy aggregation operators;
3. A formal analysis of the link between the composition of behaviors and the satisfaction of (complex) goals, based on fuzzy logic; and
4. A new solution to the problem of the integration of high-level task planning and low-level motion control, based on the automatic generation of complex combined behaviors, called B-plans.

The research methodology followed ties formal analysis and practical testing. All the techniques presented have been extensively validated in experiments run on real robots.