Position reference for autonomous mobile robots

Abstract

Sensors enable robotic systems to acquire environment data. Knowledge of the environment is essential for the performance of intelligent actions or tasks. For mobile robots, a fundamental task which must be accomplished, with the aid of sensors, is that of self-location or position determination. The objective is to minimize the position uncertainty of a mobile vehicle by utilizing vision, range, and proximity data gathered from sensors on board the vehicle. This dissertation presents an analysis of the problem and proposes models, solutions, and algorithms for robot location. Information about a robot's position is derived from geometric properties of an environment in conjunction with sensor data. We study how a-priori knowledge of the environment can provide a bound on sensor system performance. Even simple geometric properties, such as symmetry, are shown to be of importance in this investigation. Results for determining bounds on the minimum attainable uncertainty for a class of environments are given. A method is proposed, based on mathematical morphology, to determine position. The environment and sensor readings are modeled as sets and as structuring elements, respectively. Operators are defined to modify the environment representation into regions where the robot is likely to be positioned. This method can incorporate sensor noise characteristics by a proper design of the structuring elements. In the case where compass information is not available, a decision theoretic approach is taken. For a given set of sensor readings, we construct all possible sets of environment elements that could have contributed those readings. Each set is evaluated by a likelihood measure, and the most likely set to have caused the readings is selected. Once a choice is made, known techniques are employed to compute a position estimate. The position determination methods use processed sensor data such as perpendicular distances to walls or view angles of corners. The practicality of extracting such information from ranging sensors and other types of transducers is addressed, and several feature extractors for use in the self-location algorithms are shown. A feature extractor for application to range data is investigated, and results are shown for several levels of noise.