Localization Using Laser Scanning and Minimalistic Environmental Models

Patric Jensfelt

Abstract:

Abstract: This thesis deals with the problem of localization for a mobile robot in an indoor, structured and non-engineered environment. By localization we mean the process of determining the position and orientation of the robot from sensor measurements of the environment. We will experimentally show that minimalistic environmental models can be used to solve the localization problem, contrary to many other approaches where a higher degree of detail is used. The thesis begins with a brief overview of the most common sensors used for localization, including odometry, sonar, laser scanner, infrared, radar, inertial, compass and vision. We argue for the use of a laser scanner, because of its high sampling rate and excellent angular resolution. We also discuss existing localization methods and divide them into feature-based and grid-based methods. We conclude that feature-based methods are best suited for combining with minimalistic models of the environment. The first step of the localization method developed in this thesis is the derivation of models for the sensors used. The sensors used are odometry, which approximately measures the motion of the robot by counting the wheel rotations, and laser scanning. The odometry gives excellent accuracy for short distances traveled but is inherently susceptible to drift as a result of wheel slippage and modeling errors. On the other hand, the laser scanner provides lower accuracy but measures the absolute position of the robot relative to the environment. Using the sensor models and a minimalistic model of the environment we develop a method for tracking the position of the robot. We also present an algorithm for initializing the position of the robot. Both algorithms are implemented on a Nomad200 robot using a PLS laser scanner from SICK. Experimental results are presented that demonstrate robustness of the methods and show that the position can be determined with an accuracy of 20 mm, depending on the quality of the model.

BibTeX Entry: 

@TechReport{JensfeltLic99,
  author = 	 {Patric Jensfelt},
  title = 	 {Localization Using Laser Scanning and Minimalistic Environmental Models},
  institution =  {Automatic Control},
  year = 	 1999,
  type =	 {Licentiate Thesis},
  address =	 {Royal Institute of Technology, SE-100 44 Stockholm, Sweden},
  month =	 apr
}
Download: ps.gz (917k)