Perception and analysis of the environment around us is a major challenge in many promising industrial fields. In this context, artificial intelligence (AI) algorithms have unquestionably proven their effectiveness for vision-related tasks, using various sensors (camera, lidar...). Today, there is growing interest in the use of radar (radio detection and ranging) sensor data by AI. Radar is a sensor that stands out for the nature of its data, its operability (low light levels, bad weather, etc.) and its cost. However, it produces sparse data with low spatial resolution, making them difficult to exploit by traditional algorithms. Recently, artificial neural networks based on a graph representation of data (Graph Neural Networks - GNN) have shown good accuracy on sparse and noisy sensor data [1]. As a result, the use of GNNs to exploit radar data seems very promising [2]. The spectrum of applications is wide, from intelligent vehicles (cabin monitoring) to medical devices (vital signs measurement) and surveillance devices (fall detection).

[1] Dalgaty et al, « HUGNet: Hemi-Spherical Update Graph Neural Network applied to low-latency event-based optical flow » Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023, pp. 3952-3961

[2] Fent, et al., "RadarGNN: Transformation Invariant Graph Neural Network for Radar-based Perception," Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023, pp. 182-191.

Candidate profile: Final year engineering student or Master 2

Desired skills : Strong motivation to learn and contribute to artificial intelligence research. In-depth knowledge of computer science and programming languages (Python). Knowledge of artificial intelligence and experience in artificial neural networks (Pytorch or Tensorflow libraries) are a plus. The recruitment interview may refer to the two publications mentioned.