Description du projet
Concevoir des véhicules autonomes plus sûrs
Les véhicules routiers autonomes font enfin partie de notre quotidien. Les véhicules autonomes ont beau représenter un pas important vers l’avenir, ils comportent également leur lot de défis. Le principal étant est d’assurer une sécurité absolue sur les voies publiques. Les véhicules autonomes de pointe pâtissent encore d’une détection des objets médiocre et de fausses alarmes. La solution consiste souvent en l’ajout de capteurs à plus haute résolution et plus onéreux qui ne parviennent pas toujours à résoudre les problèmes inhérents aux systèmes de détection d’objets courants. Le projet STV, financé par l’UE, mettra au point une nouvelle solution de détection d’objets qui repose sur une nouvelle architecture révolutionnaire. Il vise à optimiser les capteurs bon marché, améliorer les taux de détection, réduire les fausses alarmes et concevoir des véhicules autonomes sûrs et abordables.
Objectif
The automotive industry is amid a disruptive change highlighted by the entry of autonomous vehicles. However, at current stage,
self-driving cars technologies are not safe enough for operation on public roads. They suffer from too many missed detections and
high false alarm rates. Some autonomous vehicle developers have tried to overcome these problems by putting higher resolution
(and higher cost) sensors, yet they solutions still these suffer from inadequate perception.
There is a growing market consensus that the limitations of the current perception solutions (called ‘Environmental Models’) are
entrenched in their ‘Object level’ fusion architecture. This cannot be fixed by tweaking the algorithms, changing parameters or
adding more data for learning. A promising alternative solution is ‘Raw data fusion’ with roots in academia and now diffusing to
commercial projects.
VAYAVISION “Seeing the View” project is based on ‘Raw Data Fusion’ architecture with up-sample techniques to further increase the
effective resolution of sparse measurements from active sensors (LiDARs and RADARs). The solution constructs an accurate RGBd 3D
model based even on low cost sensors while enabling the perception algorithms richer data and a more comprehensive view of the
environment. Using Machine Vision algorithms and Deep Neural Networks, VAYAVISION detects very small obstacles (such as a
10cm high box) and has much better detection rates and with less false alarms than the legacy ‘Object Fusion’ solutions.
VAYAVISION’s raw data fusion platform is planned to enable a much safer and comfortable driving experience at an affordable
vehicle price. VAYAVISION solves the heart of autonomous driving challenge of correctly understanding the changing environment
of the vehicle by using ‘Raw Data Fusion’ and Up-sampling.
Champ scientifique
- engineering and technologymechanical engineeringvehicle engineeringautomotive engineeringautonomous vehicles
- natural sciencescomputer and information sciencesartificial intelligencecomputer vision
- engineering and technologyelectrical engineering, electronic engineering, information engineeringinformation engineeringtelecommunicationsradio technologyradar
- engineering and technologyelectrical engineering, electronic engineering, information engineeringelectronic engineeringsensors
- natural sciencescomputer and information sciencesartificial intelligencecomputational intelligence
Programme(s)
Régime de financement
SME-2 - SME instrument phase 2Coordinateur
6037604 OR YEHUDA
Israël
L’entreprise s’est définie comme une PME (petite et moyenne entreprise) au moment de la signature de la convention de subvention.