PhD position - Trajectory Prediction for Autonomous Navigation

OPPORTUNITY DETAILS

Total reward

0 $

State University

CEA Tech

Area

Western Europe

Host Country

France

Deadline

31 Oct 2021

Study level

Doctoral Degree

Opportunity type

PhD

Specialities

Engineering , Information technology and computer science

Opportunity funding

Not funding

Eligible Countries

This opportunity is destined for all countries

Eligible Region

All Regions

SL-DRT-21-0644

RESEARCH FIELD

Cyber physical systems - sensors and actuators

ABSTRACT

With the growing interest in Autonomous Vehicles (AV), perception systems play a central role in their navigation, with active developments from the research and automotive industry communities. Perception systems provide AVs with information about the driving situation. Basically, advanced algorithms model the vehicle environment using a map by processing past and present data from on-board sensors such as cameras, LiDARs, radars and ultrasounds. The future evolution of the driving environment is predicted in order to plan safe trajectory, avoid collisions and make navigational decisions.CEA has developed a patented on-board sensor fusion technology that exploits the occupancy grid paradigm to model the vehicle environment. This grid provides a probabilistic estimate of occupied and free regions. The estimation of obstacle movement is also under development. However, a prediction layer that estimates the likely future trajectories of moving obstacles is still missing. The objective of the PhD thesis is to develop an embedded trajectory prediction algorithm for autonomous navigation. Trajectory prediction is a spatio-temporal (4D) problem where uncertainty is essential to evaluate the probable short-term evolution of a driving scenario. The diversity of moving obstacles makes trajectory prediction very difficult when integrated within lightweight computing platforms. In fact, a moving car does not have the same degree of freedom as a pedestrian. Prediction models can take into account the nature of moving obstacles if this information is available (for example, provided by artificial intelligence). Otherwise, prediction models must adapt to the available data. During the thesis, the PhD student will first focus on the probabilistic modeling of motion and trajectory. Then, he/she will propose a low-complexity algorithmic solution that can run in real time on an embedded computing platform. The PhD student will be hosted in a team whose expertise is the development of advanced and lightweight perception solutions that can be integrated into embedded systems. The PhD student will collaborate with researchers, engineers and other PhD students from various scientific fields. The candidate must have a strong mathematical background in probability/statistics, computer science and software prototyping (matlab/python, C++). Knowledge and skills in artificial intelligence and data fusion will be a plus.