1 Benchmarking Robot Path Planning

The University of Tulsa, among other research institutions, performs research exploring the possibility of using Genetic Algorithms (GA’s) to develop an algorithm for autonomous robot navigation. We have achieved significant success in this area. However, we encountered a problem. We realized there is no standard metric for comparing different algorithms developed by many researchers working in this field. Our most recent work involves developing a benchmarking system to evaluate different GA’s. The concept behind our benchmark program is to provide a set of standard paths that have been mathematically evaluated and classified in different categories as Easy, Moderate, Difficult and Very Difficult. INTRODUCTION An autonomous robot performs without any external control from a human operator. The main components involved in the operation of an autonomous robot are visual detection of the environment, path planning and controls. Path planning has been defined as determining if a continuous and obstacle-avoidance sequence of positions and orientations of the robot exists, starting with the initial position and orientation of the robot, and ending with the goal position and orientation. To date, several path-planning algorithms have been developed to achieve this end using several techniques from neural networks and fuzzy logic to genetic algorithms (Cazangi and Figuieredo 2002; Di Gesu et al. 2000; Hocaoglu and Sanderson 2001). However, there has been no standard for the comparison of the performance of these algorithms. Hence, the objective of our research is to define and formulate a metric for defining the complexity of an obstacle-filled navigation area, or search space, and use this metric to develop a mathematical rating system to classify the search spaces. This metric, the complexity index, provides a way to evaluate and classify search spaces. The aim is also to create a standard set of search space patterns with varying complexity index