Chercheur Doctorant

**Scientific context**:
The 4th industrial revolution is strongly affected by the massive use of the emerging information and communication technologies such as Robotics, Artificial Intelligence (AI) and eXtended Reality (XR) in pair with the Digital Twin (DT), defined as the high-fidelity virtual model of the physical system in virtual space. One current challenge is to keep the DT up-to-date according to the evolutions of the physical system so as the supervisor decision-making would be the most relevant for enhancing productivity, making individualized and lower-cost products, strengthening the operator’s safety and efficiency and making better use of available resources (Huang et al., 2021). To achieve this physical to virtual updating, one solution consists in taking advantage of the robot perception capabilities by detecting and estimating the 6D pose ([,,], [,,]) of industrial objects disseminated in the warehouse.

6D pose estimation of rigid objects is a well-known problem in computer vision, and several relevant surveys have already been published (Marchand, Uchiyama and Spindler, 2016), (Chen et al., 2020). Current approaches can be classified in 3 main categories : 1) Traditional methods based on: point-pair features; template matching (Hinterstoisser et al., 2011); 3D local features (Buch, Kiforenko and Kraft, 2017); 2) Deep Learning methods based on Convolutional Neural Networks (CNNs): SSD-6D (Kehl et al., 2017); PoseCNN (Xiang et al., 2018); Deep6DPose (Do et al., 2018); 3) Mixed approaches combining traditional and deep learning methods: BB8 (Rad and Lepetit, 2017); YOLO-6D (Tekin, Sinha and Fua, 2018) and YOLO6D+ (Kang et al., 2020).

**PhD Proposal**:
The contributions of this PhD proposal tackle the 6D pose estimation of texture-less objects in an industrial context, and aim to revisit Pitteri's approach by questioning possible alternatives to LSE descriptors and its processing architecture to make the 6D pose estimation real-time.

Two main work avenues are envisaged:

- One focuses on the use of Neural Networks (NN). Despite promising results carried out over textured objects, our contributions will rely on evaluating the performances (pose estimation errors, execution time) and extending the approach for managing with texture-less objects and making it real-time.
- Another research avenue will rely on considering 3D-2D surface-based approach (instead of dense vertices). Since the pose estimation will be estimated up to a scale factor, we will constrain the problem by considering the geometry of the 3D model. Specific consideration will be given to the mesh refinement of the CAD model in terms of number and size of surfaces, in order to find a relevant compromise between pose estimation performances, quality of the CAD model and execution time.

**Laboratory overview**:
The objective of research at CESI is to produce knowledge and methods at the service of the scientific community. Its current organization is linked to the fields of training and helps a better identification of CESI LINEACT (EA 7527) within the research ecosystem. It puts its results at the service of CESI’s partner companies and contributes, in its fields, to take up the technological challenges in the territories that host and nourish it. It is also at the service of CESI learners to participate in their training through research.

CESI‘s historical proximity to companies is a determining factor for our research activities, and has led us to focus our efforts on applied research close to the company and in partnership with them. A human-centered approach coupled with the use of technologies, as well as the territorial network and the links with training, have made it possible to build a transversal research; it puts the human, its needs and its uses, at the center of its problems and approaches the technological angle through these contributions.

**Your skills**:

- Team working
- To be rigorous
- Writing skills
- Expected scientific and technical skills

**To convince you a little more:
- Executive fixed term contract
- Remuneration over 13 months
- 6 weeks of paid vacation
- 14 days of RTT
- Teleworking Charter
- Laptop

And above all a good atmosphere guaranteed

If this profile suits you and you share CESI's values.