Localisation of a Mobile Robot using Laser Scanner and Reconstructed 3D Models
João Gomes Mota
Thesis to obtain Master of Science degree in Electrical Engineering, Instituto Superior Técnico, November 2001.
This thesis describes in detail the localisation solution I developed for the RESOLV project. During the project life, the operation range of the RESOLV system extedend to new and more challenging environments and the localisation solution adapted to the new scenes, while keeping a high degree of robustness and low estimation error..
- Mobile Robot Localisation
- Laser Scanner
- 3D Scene Reconstruction
- Frame Localisation
Members of the Jury
The thesis is a 216 page document, written in english. There is an extended abstract in Portuguese (Adobe® PDF, 25 pages), which provides a good introduction.
A thesis is available in Adobe® PDF format, divided in seven documents:
- Prologue (abstracts, acknowledgments and table of contents)
- The Frame Localisation Algorithm
- The Likelihood Test
- The Approximate Localisation Algorithm
- References and Appendices
This dissertation presents a novel procedure for the localisation of a mobile ro-bot, supported on distance profiles acquired with a laser range scanner. The localisation system is embarked on a mobile robot, whose main task is to acquire and reconstruct a 3D map of large indoor environments.
No prior knowledge or a priori constraints of the environment map are known or assumed for localisation purposes. At each stage of the reconstruction procedure the environment map is known up to the extent it has already been reconstructed.
The main contributions of this dissertation are two algorithms and one dedi-cated statistics. The first algorithm, named Frame Localisation, is based on natural feature extraction. It uses no a priori estimate of the robot's position and produces a reasonably good position estimate. The second algorithm, named Approximate Lo-calisation, provides a refined position estimate based on an initial estimate or the results of the first algorithm. It compares the scene's range profile with the
recon-structed map and computes a maximum likelihood fit. The statistics measures the quality of the computed posture estimates, providing a quantitative performance criterion.
This dissertation presents experimental relevant results, obtained during real 3D acquisition campaigns on different types of indoor environments.