This paper presents the design and provides a partial analysis of the performance of an autonomous ground robotic vehicle called Vasilius. Applications for Vasilius include autonomous navigation on a somewhat marked path with obstacles, leader following, and waypoint navigation. The paper focuses on three aspects of Vasilius: the design, the performance, and a technique for filtering, mapping, and learning. The design of Vasilius embodies a novel idea of modeling an autonomous vehicle after human senses and the human decision-making process. For instance, Vasilius integrates information from seven types of independent sensors, and categorizes them into either short-range reaction sensors and/or long-range planning sensors, analogous to what the human brain does. The paper also analyzes the performance of Vasilius, relating theoretical predictions to actual behavior. Some of these analyses, especially for the filtering, mapping, and learning, are still in progress. Performance measures that have been measured include speed, ramp climbing, turn reaction time, battery life, stop reaction time, object detection, and way-point accuracy. Finally, the paper discusses Vasilius' use of a new approach to filtering, mapping, and learning to enhance its performance.
展开▼
