Robotics and Autonomous Systems 54 (2006) 277–287 www.elsevier.com/locate/robot Feature extraction for outdoor mobile robot navigation based on a modified Gauss–Newton optimization approach Sen Zhang ∗ , Lihua Xie, Martin David Adams School of Electrical and Electronic Engineering, BLK S2, Nanyang Technological University, Singapore 639798, Singapore Received 11 February 2004; received in revised form 9 November 2005; accepted 16 November 2005 Available online 18 January 2006 Abstract This paper discusses the problem of feature detection for semi-structured outdoor environments such as campuses and parks using laser range sensors. In these environments, commonly encountered natural features that can be very useful for mobile robot navigation include edges (large discontinuity) and circles (e.g., trees, pillars). The term feature is used to denote objects which are “likely” to be detectable when the sensor is moved to new locations. Note that there has been no systematic approach for feature detection in outdoor environments. In this paper, we present an algorithm for feature detection. The algorithm consists of data segmentation and parameter acquisition. A modified Gauss–Newton method is proposed for fitting circle parameters iteratively. Experimental results show that the proposed algorithm is efficient in detecting features for semi-structured outdoor environments and is applicable to real time simultaneous localization and mapping. c  2005 Elsevier B.V. All rights reserved. Keywords: Semi-structured outdoor environments; Feature extraction; Data association; Optimization; Mobile robot 1. Introduction Navigation is one of the basic problems for autonomous mobile robots. Its history can be traced back to 4000 years ago. Today, navigation is a well-understood quantitative science, used routinely in maritime and aviation applications [15,2, 12]. Given this, the question must be asked as to why robust and reliable autonomous mobile robot navigation remains such a difficult problem. The core of the problem is the reliable acquisition or extraction of information about navigation beacons from sensor information and the automatic correlation or correspondence of these with some navigation map [7,11]. Lots of navigation systems use artificial beacons to realize their navigation task, but the approach may not be realistic in applications such as exploration of jungles or other unknown environments. In this situation, one needs to utilize naturally occurring structures of typical environments to achieve a similar performance. Hence, fast and reliable algorithms capable of extracting features from a large set of noisy data ∗ Corresponding author. E-mail addresses: ps7184247i@ntu.edu.sg (S. Zhang), elhxie@ntu.edu.sg (L. Xie). are important in such applications. Some of the early efforts in this direction have focused on extracting line features in an indoor environment based on the information provided by sonar and laser sensors. In Ref. [5], a least-squares line fitting technique was applied to extract edges from ultrasonic sensor data. In Ref. [18], a recursive line fitting system is used to extract line segments under polar coordinates and an ellipse fitting method is also implemented for data from a laser sensor. In Ref. [19], line segments are detected using a regression least-squares parameter estimation method whereas the center and radius of a circle feature are estimated based on the average value of the measurements of the circle from a 2D range scanner. Instead of fitting straight line segments after a full scan has been recorded, Adams presented an on- line edge extraction approach employing a Kalman filter in Ref. [3]. Later, based on this method, a two-layer Kalman filter is used to calculate the parameters of a line by on-line means in Ref. [17]. Observe that the aforementioned works are focused on indoor applications and are mainly concerned with line extraction. For an outdoor environment, the problem of feature selection and detection is more challenging. In our view, in most typical semi-structured outdoor environments, such as campuses, parks 0921-8890/$ - see front matter c  2005 Elsevier B.V. All rights reserved. doi:10.1016/j.robot.2005.11.008