トゥールーズのISAE-Supaeroがロボット工学や人工知能に関心を持つ博士課程(経済援助あり)の学生を募集しています。

The PhD student will be situated in the DEOS research department , SCAN Team (Signal communication Antenna Navigation) in ISAE-Supaero Campus, Toulouse, France. The PhD subject is part of the research activities of SCAN on environment perception for autonomous vehicle navigation and multi-sensors data fusion. The research work started many years ago with a strong interest in the increasing of GNSS positioning accuracy in harsh environments. Today, with the strong development in the team of visual navigation technics, this thematic is evolving in the framework of the ANR AVISE project with « Semantic
Visual Analyze ».

PhD Subject

Current research works in the field of autonomous navigation focus mainly in the study of localisation and mapping algorithms based on hybridization of multiple sensors. Today, some efficient technic works such as ORB-SLAM (Mur-Artal 2015), SVO (Forster 2016), PTAM (Klein 2009). All these methods can be considered as « low-level » approaches as the scene understanding is very limited. Indeed the scene is represented as a 3D point cloud without any semantic information.

Let’s note that with machine learning and Deep Learning, some scene analysis technics are emerging with object detection and recognition (pedestrian detection (Benenson 2014), traffic signs, zebras (Soheilian 2010-2013)). Nevertheless, such approaches are still decorrelated from the navigation step.

The ambition of this project is to integrate the scene analyse step in the framework of the autonomous navigation, that means to integrate semantic information in the position processing step.

We aim to build a semantic urban object mapping (traffic signs, traffic lights, zebras, shops...) and event mapping (works, deviation, crashes…). Such a mapping will allow to based the navigation on high level landmarks more robust in time and in environment changing conditions (day, night, rain, fog...)

 

This project is at the intersection of multiple fields:

  • Automatic machine learning, image understanding and object detection
  • Vision based localisation (visual odometry, hybridization)
  • Geolocalized semantic mapping (SLAM+GNSS)

 

The main challenges to tackle are :

1 – To build a vision based system to understand the environment of the vehicle using multi-sensor data (Camera, IMU, GNSS (GPS/Galileo). The objective is to detect and localize semantic objects in the surroundings of the vehicle,

2 – To build a mapping of the detected semantic objects is such a way that the map could be used and enriched in the case of multi-vehicle collaboration. An effort will also be done to analyse the performance and integrity of such positioning and mapping approach.

 

Profile of the candidate

The candidate must demonstrate an interest in the subject and have a strong scientific curiosity,

The Candidate should have an MSc or equivalent in Science and/or Technology with some competencies in one or several of the following topics:

- computer vision (epipolar geometry, visual odometry, SLAM, SfM, 3D reconstruction, ...)

- Machine Learning (Neural netwok, SVM, KNN, Random Forest …)

Skills in Matlab, C / C ++, OpenCV, ROS would be very appreciated,

The French language level is not a requirement (French classes will be proposed during the PhD).

A good English level is requiered.

 

Applications

Applications must be submitted to Dr Damien Vivet (damien.vivet@isae.fr). Before 31st september 2017.Applications should include a detailed CV, master grades, ranking and marks sheets, a covering letter and