Teoresi è una società internazionale di servizi di ingegneria, nata a Torino nel 1987.
Siamo specializzati nel supportare le aziende nella realizzazione di progetti che utilizzano tecnologie all'avanguardia, dalla guida autonoma alle nanotecnologie applicate all’ambito medicale. Il nostro approccio innovativo prevede una stretta collaborazione con i reparti di Ricerca e Sviluppo dei principali marchi industriali. Realizziamo soluzioni chiavi in mano accelerando il time-to-market del cliente. Teoresi è una delle 10 aziende selezionate da Amazon per collaborare allo sviluppo di nuovi prodotti basati sull’interazione vocale di Alexa.
Siamo sempre alla ricerca di persone di talento da inserire nel nostro team. In Teoresi diamo valore agli aspetti innovativi di ogni sfida progettuale , al lavoro di squadra, alla diversità e e ci piace pensare liberi da confini, non solo geografici.
Siamo costantemente aggiornati sui progressi tecnologici, dando priorità alle persone e alla sostenibilità ambientale. Il nostro team multidisciplinare e la nostra presenza globale ci permettono di offrire opportunità di carriera internazionali e di soddisfare le esigenze di un mercato in costante evoluzione. Crediamo che la proattività e la curiosità per l'apprendimento continuo siano essenziali in un contesto di squadra e ci impegniamo a generare innovazione in tutto ciò che facciamo.
Se condividi i nostri valori e ti interessa far parte di un'azienda orientata al futuro, continua a leggere e candidati!
Description
Teoresi is looking for talented young people interested to develop a thesis project in the company.
In this case we introduces a master thesis proposal that focus on Artificial Intelligence Algorithm based on StereoVision using a Neural Processing Unit board.
Company Teoresi Group -> Teoresi S.p.A. | Italy
Job requirements
Topic Characteristics:
ADAS and autonomous driving cover many different technologies: vision and artificial intelligent algorithm, embedded high efficient C++ code, planning algorithms, new control algorithms, simulation and unity assets design, Navigation, Human-Machine-Interface, Connectivity.
In the context of Autonomous driving, neural networks are crucial for tasks such as lane detection, object detection, image segmentation etc. Currently neural networks are not optimized for embedded target. Only recently, optimized Hardware is available on the market for a real and efficiently porting. In particular NVidia with the latest AGX Xavier, implements internally two neural processing unit in order to optimize neural network inference.
This thesis proposal focus on implement a state-of-the-art neural network for Stereo-Vision cameras in order to better estimate distance between objects. Sensor Fusion module estimate distance using Lidar and Radar, but also cameras can provide data in order to better analyze the environment and provide an additional point of view in order to perform a plausibility check of the reconstruction. Autonomously identify different bottleneck on deployment and analyze performance and memory occupation on the target. The Implementation will be integrated in a photorealistic simulator with the aim of bringing it to a real target.
Methodology:
The goal of this work is defined by these step:
- Stereo-Vision Algorithms Study and identification:
o The first phase is focus in the identification of the better memory/performance trade off algorithm for stereovision distance estimation based on vision techniques and Neural Network.
- Implementation:
o Development of C++ code for stereovision algorithm included stereovision synchronization check on a real board with dual camera. A first functional implementation is required to test the accuracy of the network based on the state of the art results. This is the baseline for the next optimization step.
- CNN optimization:
o To optimize embedded porting of the algorithm is important write specific CUDA kernels in order to speed up the execution exploiting the GPU inside the board. TensorRT can also be used if the identified algorithm is a Neural Network.
- CNN Simulation on a real Autonomous driving framework
o The test bench consist of a series of test use case on a real urban scenario inside a Unity base Simulator. The performance of the neural network must be compatible with the needed of a real car reaction.
- Test on a real Target
o The last step is the test of the algorithm in a real scenario with a real autonomous vehicle
Toolchain:
Tensorflow/Caffe
TensorRT
Unity Engine
CUDA
C++
Preferential position requirements
At the end of the project, and after the graduation, there could be the opportunity to be included in the company, through a paid internship.
Interested parties are invited to respond to the following add by uploading their CV, or to send an email to job@teoresigroup.com, referring to the add (TeoTESI_AISV_201904) and attaching the detailed CV with the course of study and exams taken.
The opportunity offered is to work in a young and dynamic environment, able to recognize and reward the best professionals.
Under current legislation, the job offer is intended to be extended to both sexes (Law 903/77).
Education
Laurea tradizionale o specialistica
Career level
Laureando / neolaureato
