My PhD and the two following years of Post-Doctoral position

I prepared my PhD thesis in the ICARE project-team of INRIA Sophia-Antipolis. It was titled "Robust localization of a vehicle in urban environment with a stereo-vision system ", my supervisor was Patrick Rives.

My PhD was funded by the CyberCars consortium, which aims to define the future transportation systems in downtown areas by introducing the concept of vehicle-sharing. Indeed, due to their expansion, high density cities might not allow a private car for each inhabitant. The European Communauty has decided to group the main european laboratories in robotics to develop the Cybercar, a mobile robot which will improve the quality of life in our urban environments. The final goal was the creation of an automatic, safe, silent and environmently friendly vehicle which will represent a new alternative between the public transports and the non-motorized ones.

Most of us have already seen unmanned vehicles running along highways or restricted roads. In fact, these experimentations represent the civil applications of all the recent research on mobile robots. In outdoor applications, all mobile robots use a GPS system to locate their pose. To obtain reliable GPS data, the receptors require free space that can rarely be found in urban environments: the facades generally obstruct the reception of satellite signals.

My work consisted of finding a solution to the loss of GPS data in streets to finally locate a mobile robot. The urban environment is certainly one of the most complex environments to locate a mobile robot. The complexity of such scenes comes from the number of dynamical elements (vehicles, pedestrians) and the technical and financial impossibility of using dead-reckogning systems for long path following.

I had thus been developing since october 2001 a vision-based method. Vision systems are particularly weel-adapted to my application and do not require the modification of the environment.

The main assumptions I made are:
  1. the road is considered as locally planar and have parallel boundaries,
  2. the high-frame rate of the camera induces a small motion of the features,
  3. the image foreground (first meters before the vehicle) is generally free from obstacles.
The proposed method allows an estimation of the motion of a vehicle in the difficult conditions of urban traffic. The use of a stereo rig allows us to reject most features which do not lie on the road plane. The algorithm can cope with most obstructions of the field of view for example the crossing of pedestrians.

I joined the Centre de Robotique of Ecole des Mines de Paris in september 2005 to pursue my thesis works on Computer Vision dedicated to Intelligent Transport Systems. I have been spliting my time between Ecole des Mines and INRIA which formed in december 2005 the Joint Research Unit La Route Automatisée. For september 2006 to august 2007, I pursued my works in IMARA of INRIA Paris/Rocquencourt.