Note: This content is accessible to all versions of every browser. However, this browser does not seem to support current Web standards, preventing the display of our site's design details.


Navigation of autonomous vehicles

Navigation of autonomous vehicles can be divided into four principal categories: perception, localization and map building, motion planning and motion control. Perception is concerned with extracting relevant information from the environment using various proprioceptive and exteroceptive sensors. An exteroceptive sensor commonly used is vision sensor that is applied here to lane detection for a driver assistance navigation module in personal cars. A set of image filters is tested within a probabilistic framework based on particle filtering where the state of the system is represented by a discrete set of particles each representing a possible belief on the relative vehicle position with respect to lanes. Thus inferred solution proves robust under various road conditions. Motion planning provides an autonomous vehicle with a feasible path from a starting configuration to a goal point. An on-line path planning technique based on grid map representation of a dynamic environment is the D* graph search algorithm generating the shortest path solution that can be further adjusted locally using the free-space bubble concept. Off-line path planning can provide different criteria of optimality on the global path such as path smoothness and free-space safety that is achieved here by B-spline curve optimization on its envelope with respect to the polygonal channel representation of the environment. Motion control at the reactive level provides motion commands based only on current sensory information for local obstacle avoidance. An integration of the dynamic window reactive approach to obstacle avoidance that takes into account also the global path connectivity will be presented. A global path following technique called modified virtual vehicle that generates motion commands based on the position and orientation error feedback to a reference point will be furthermore presented. Both motion control algorithms are validated on a differential drive mobile robot. Finally, a reactive level learning agent that develops a motion control strategy based on weak reinforcements from the environment will be exposed.
Type of Seminar:
Public Seminar
Kristijan Macek
ASL Autonomous Systems Lab, EPFL, Lausanne
Dec 01, 2004   17:15

ETH-Zentrum, Gloriastrasse 35, Zurich, Building ETZ , Room E6
Contact Person:

Prof. M.Morari
File Download:

Request a copy of this publication.
Biographical Sketch:
1994 International Baccalaureate, Maribor// 1999 Diploma degree in Electrical Engineering with title “Application of fuzzy-neural networks in identification and control”, Faculty of Electrical Engineering and Computing, Zagreb// 1999 Internship at Department of Electronics, University of Barcelona// 2000-2003 Research assistant at the Department of Control and Computer Engineering in Automation, Zagreb// 2003-2004 Visiting student at Autonomous Systems Lab, Ecole Polytechnique Fédérale, Lausanne// 2004 M.Sc. degree in Electrical Engineering with title “Motion planning of mobile robots in indoor environments”, Faculty of Electrical Engineering and Computing, Zagreb. His current research focus has been in navigation of autonomous vehicles.