Description
Date depot: 20 mars 2019
Titre: Multi-sensors Raw-data Fusion for Next Generation Autonomous Driving Systems
Directeur de thèse:
Pietro MICHIARDI (Eurecom)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé:
Tomorrow's mobility is progressively
bending towards partial or full autonomy of vehicles. Many efforts are provided
by the research and industrial community to make Autonomous Driving (AD) a
reality. One of the scientific and technical challenges associated to the
autonomous driving problem is the liability and robustness in understanding the
surrounding environment from the measurements acquired by the vehicle embedded
sensors. These perception systems (cameras, RADARs, LIDARs,
Ultra-Sonic-Sensors, etc.) require a variety of down-stream technologies to
detect, localize and semantically label agents and obstacles around the
autonomous vehicle.
In the context of this Ph.D. Thesis
project, the focus will be on the design, analysis and evaluation of
probabilistic data fusion techniques with the ultimate goal of building an
ego-centric world model to be used by downstream components in charge of using
it to take driving decisions. The challenges we will address in this thesis
revolve around: 1) efficient definition and computation of an appropriate
multi-dimensional feature space to optimally exploit the data acquired from
multiple sensor types (cameras, RADARs, LIDARs, Ultra-Sonic-Sensors, etc.); 2)
probabilistic statistical models for classification to detect and understand
the surrounding objects from the computed multi-dimensional feature space; 3)
quantification of the uncertainty affecting the detection problem and
interpretability of the results in terms of performance applicable to the
automotive field.
This Thesis is supported by Renault Software
Labs, which will prove fundamental in sharing data collected from a variety of
sensors, in addition to several thousands of kilometres of (possibly labelled)
road driving data.
Doctorant.e: Da Silva--Filarder Matthieu