Company

AMPERE SOFTWARE TECHNOLOGY

Job Description

Thesis subject description

The introduction of advanced driver-assistance systems (ADAS) has considerably enhanced the driving experience by providing automation of driving tasks, like lane keeping assist, contextual cruise control, automated emergency braking, etc. Safety and driving comfort are at the center of these changes. The reliability of ADAS features strongly relies on the robustness of the perception system beneath it. Perception system fuses information from different sensors (e.g., front camera, front radar, side radars, sonar, …) in order to obtain a reliable description of the world environment.

 

While the performances of ADAS features (and thus, the perception system) are largely validated in standard conditions (e.g., good weather conditions, clean and brand-new sensors, standard driving cases, …), it is clear that a degradation in performances is expected in difficult conditions (e.g., at night or heavy rain/snow, …). That is why each sensor must output a degradation level so that in critical conditions, the ADAS feature can simply be deactivated if required.

 

In this thesis, we want to gain robustness and accuracy of Renault’s perception system in difficult conditions by auto-adapting the sensor fusion algorithm to the conditions. After identifying a set of factors that considerably degrade performances of the current perception system, the research task consists in auto-adapting part of the perception system (e.g., tuning parameters, cost functions, …) based on these factors using self-learning. This adaptivity will then allow new algorithmic enhancements of core fusion algorithm that would not have been possible without the adaptivity and self-learning mechanisms. Several challenges will arise: convergence of the learning procedure, non-regression on standard use cases, overfitting on environmental conditions, centralized versus distributed learning, etc…  All these challenges will establish a complete research topic. The goal is to bring reliability of ADAS features even in difficult environmental conditions.

 

Your missions

A PhD thesis allows you to develop several skills to enable you to carry independent research and, at the same time, to develop the know-how on key technologies.  You shall have the opportunity to formulate novel solutions whilst making the most of our prototype platforms and driving data. You will need to gain a strong understanding of sensor data fusion and principles of machine learning & deep learning from the autonomous drive perspective and gain critical thinking to formulate the problem, propose solutions, and test them. 

 

Your profile

You should be completing or have Bac+5-level degree: an engineering diploma or a Master of Science in Engineering (Electrical Engineering, Signal Processing, Computer Science, Machine Learning, …). You are very much interested in perception systems and sensor data fusion along with self-learning algorithms (machine learning and deep learning). It is very important to be curious, willing to learn new techniques. You will have the opportunity to formulate your own ideas, to test them on Renault driving database or in Renault prototype vehicles. Strong technical background in signal processing, fundamentals of machine learning & deep learning, programming (C, Python) and scripting skills are required.

A working knowledge of the English language is required. You will have to interact with team members and several Renault divisions, at different levels.

Vous vous reconnaissez dans cette annonce ? Alors ce poste est fait pour vous!

Rejoignez-nous ! #CarMakerCareMakers

Job Family

Transverse

Contract Duration

36 months