Terrain-aided localization of autonomous ground vehicles $ R. Madhavan a, * , H.F. Durrant-Whyte b a Intelligent Systems Division, National Institute of Standards and Technology, Gaithersburg, MD 20899-8230, USA b Australian Centre for Field Robotics, The University of Sydney, Sydney, NSW 2006, Australia Abstract The development of a terrain-aided localization framework for autonomous ground vehicles (AGV) operating at high speeds in unstructured, expansive and harsh environments is the subject of this article. The localization framework developed is sufficiently generic to be used on a variety of other autonomous ground vehicles and is demonstrated by its implementation using field data collected from two different trials on two different vehicles. The results demonstrate the robustness of the proposed localization framework in producing reliable and accurate position estimates for autonomous vehicles operating in a variety of unstructured domains. D 2003 Elsevier B.V. All rights reserved. Keywords: Outdoor localization; Map building; Iterative closest point; Extended Kalman filter; Bayes theorem; Entropy; Scale space 1. Introduction The research addressed in this article is concerned with the theoretical development and practical imple- mentation of reliable and robust localization algo- rithms for autonomous ground vehicles (AGV) operating at high speeds in unstructured, expansive and harsh environments. Localization is the ability of a vehicle to determine its position and orientation (pose) within an operating environment at any given time. The need for such a localization system is motivated by the requirement of developing autono- mous vehicles in applications such as mining, agri- culture, cargo handling and construction. The main drivers in these applications are safety, efficiency and productivity. The approach taken to the localization problem in this article guarantees that the safety and reliability requirements imposed by such applications are achieved [9,12]. The problem of autonomous vehicle localization has received considerable attention from roboticists in recent years and many methods have been proposed. These methods vary significantly, depending on the environment in which the vehicle is to navigate and the types of sensors that are available. A survey of the vast body of literature [1 – 4,6,17] on robot localiza- tion indicates that the proposed approaches can be broadly classified as relative and absolute methods. Most implementations of localization algorithms com- bine the above two methods. 0926-5805/$ - see front matter D 2003 Elsevier B.V. All rights reserved. doi:10.1016/j.autcon.2003.08.006 $ Commercial equipment and materials are identified in this article in order to adequately specify certain procedures. Such identification neither implies recommendation or endorsement by the National Institute of Standards and Technology, nor does it imply that the materials or equipment identified are necessarily the best available for the purpose. * Corresponding author. Tel.: +1-301-975-2865; fax: +1-301- 990-9688. E-mail addresses: raj.madhavan@nist.gov (R. Madhavan), hugh@acfr.usyd.edu.au (H.F. Durrant-Whyte). www.elsevier.com/locate/autcon Automation in Construction 13 (2004) 83 – 100