Optimal Path Planning Strategies and Assistance in - Paris, France - ECE PARIS ECOLE D'INGÉNIEURS

Description

Optimal Path Planning Strategies and Assistance in Autonomous Driving:

• Réf

-
ABG-115915

• Sujet de Thèse 19/07/2023
• Autre financement privé
• ECE PARIS ECOLE D'INGÉNIEURS
• Lieu de travail
• Paris
• IledeFrance
• France
• Intitulé du sujet
• Optimal Path Planning Strategies and Assistance in Autonomous Driving
• Champs scientifiques
• Informatique
• Robotique
• Mots clés autonomous vehicle, machine learning, intelligent transportation, modelization, optimization
  Description du sujet:

Introduction:
An autonomous vehicle is an intelligent vehicle that is capable of driving automatically,
independent and is capable of making decisions concerning real traffic situations and related
infrastructure with or without intervention from a human being. If a human intervenes, it is
generally limited to supervision.

Autonomous vehicles are useful in many ways - for driverless travel, to accomplish some risky
military missions without loss of human lives, to help the aged and the disabled, and the likes.
measures that detect danger and obstacles and suggest/correct inappropriate driver actions
through its on-board driver assistance systems (ADAS). In addition, autonomous vehicles provide
efficient energy consumption, and user comfort.

The development of autonomous vehicles is expected to bring several economic, societal, and
environmental benefits. This, however, entails various scientific and technological challenges. To
begin with, autonomous vehicles require four fundamental steps: localization, perception,
trajectory planning and control of the vehicle.

• Localization denotes the exact position of the vehicle, where it is and where it is heading.

Tracking the vehicle, no matter where it is and what weather conditions exist, is essential in
- controlled operation of an autonomous vehicle.

• The perception of the environment entails the detection of the road, the lanes, the entities

around the vehicle, including fixed and mobile obstacles, the weather, and any incident
- happening within the vicinity near the vehicle. The perception process uses computer vision
- in which various sensors are used to achieve such objective. Among the sensors used are the
- camera, lidar, radar, GPS, and UAV (drone).Perception makes it possible to generate a dynamic map of the environment close to the
vehicle. Localization allows the vehicle to locate itself in the global reference, which would
make it possible to define the routes to follow to reach its navigation objective.

• The trajectory planning generation consists of calculating a reference trajectory, which

avoids obstacles, prevents accident, and respect other criteria, such as user comfort,
- passenger safety and adherence to traffic rules and regulations.

• Vehicle control relates to maneuvering the vehicle, using actuators, such as the steering

wheel, the brake and accelerator, to follow the reference trajectory. This step is composed of
- two steps: longitudinal control, and lateral control

Objectives:
The objective of this thesis is to design a plan for an autonomous vehicle so that it can circulate
safely and without collision in an urban environment setting that includes the presence of other
moving obstacles.

Various automatic trajectory planning techniques have already been proposed in the literature,
but none has really stand out as a general method that can satisfactorily solve the planning
problem (find a trajectory, minimize the cost function), in particular in the industrial level. As a
first approximation, we can distinguish two types of methods: those which seek to construct a
representation of free space (global methods) and those which are based on local information to
incrementally construct a trajectory (local methods).

In this context, we propose that the solution is focused on the local methods. We are interested in
presenting a new planning method which is based on artificial intelligence techniques (machine
learning) initially and then the implementation of software solutions in hardware solutions using
FPGA or similar scheme. In this work, the path is defined from the modeling of the environment
of the autonomous vehicle using neural networks. Such path is equivalent to a cost function which
remains to be formalized, considering some constraints such as energy reduction, improved
comfort, and road safety. The complete path will be a composition of a set of paths, each of which
has its own cost function.


Prise de fonction:

• 01/11/2023
  Nature du financement:
• Autre financement privé

Précisions sur le financement:

• Projet financé par ECE Ecole d'Ingénieurs
  Présentation établissement et labo d'accueil:
• ECE PARIS ECOLE D'INGÉNIEURS

L'ECE (École centrale d'électronique) est l'une des 204 écoles d'ingénieurs françaises accréditées au 1er septembre 2020 à délivrer un diplôme d'ingénieur4.

Ses locaux sont situés dans le 15e arrondissement de Paris et dans le 7e arrondissement de Lyon.

Elle form