Exploration of unsupervised approaches for modeling the environment from RADAR data
Radar technologies have gained significant interest in recent years, particularly with the emergence of MIMO radars and "Imaging Radars 4D". This new generation of radar offers both opportunities and challenges for the development of perception algorithms. Traditional algorithms such as FFT, CFAR, and DOA are effective for detecting moving targets, but the generated point clouds are still too sparse for precise environment model. This is a critical issue for autonomous vehicles and robotics.
This thesis proposes to explore unsupervised Machine Learning techniques to improve environment model from radar data. The objective is to produce a richer model of the environment to enhance data density and scene description, while controlling computational costs for real-time computing. The thesis will address the question of which types of radar data are best suited as inputs for algorithms and for representing the environment. The candidate will need to explore non-supervised algorithmic solutions and seek computational optimizations to make these solutions compatible with real-time execution.
Ultimately, these solutions must be designed to be embedded as close as possible to the sensor, allowing them to be executed on constrained targets.
CORTEX: Container Orchestration for Real-Time, Embedded/edge, miXed-critical applications
This PhD proposal will develop a container orchestration scheme for real-time applications, deployed on a continuum of heterogeneous computing resources in the embedded-edge-cloud space, with a specific focus on applications that require real-time guarantees.
Applications, from autonomous vehicles, environment monitoring, or industrial automation, applications traditionally require high predictability with real-time guarantees, but they increasingly ask for more runtime flexibility as well as a minimization of their overall environmental footprint.
For these applications, a novel adaptive runtime strategy is required that can optimize dynamically at runtime the deployment of software payloads on hardware nodes, with a mixed-critical objective that combines real-time guarantees with the minimization of the environmental footprint.
Securing Against Side-Channel Attacks by Combining Lightweight Software Countermeasures
Side-channel attacks, such as analyzing a processor's electrical consumption or electromagnetic emissions, allow for the recovery of sensitive information, including cryptographic keys. These attacks are particularly effective and pose a serious threat to the security of embedded systems.
This thesis focuses on combining low-impact software countermeasures to strengthen security against side-channel attacks, an idea that remains poorly explored in the current state of the art. The goal is to identify synergies and incompatibilities between these countermeasures to create more effective and lightweight solutions. In particular, low-entropy masking countermeasures will be considered.
These ideas will be applied on cryptography algorithm, with a particular focus on post-quantum cryptography algorithms.
The thesis aims to develop new ways to secure software, offering better trade-offs between security and performance than existing approaches.
Efficient Multimodal Vision Transformers for Embedded System
The proposed thesis focuses on the optimization of multimodal vision transformers (ViT) for panoptic object segmentation, exploring two main directions. The first is to develop a versatile fusion pipeline to integrate multimodal data (RGB, IR, depth, events, point clouds) by leveraging inter-modal alignment relationships. The second is to investigate an approach combining pruning and mixed-precision quantization. The overall goal is to design lightweight multimodal ViT models, tailored to the constraints of embedded systems, while optimizing their performance and reducing computational complexity.