As an intern at the CEA, you will have the opportunity to work in a world-renowned research environment. Our teams consist of passionate and dedicated experts, providing an environment conducive to learning and collaboration. You will have access to state-of-the-art equipment and top-tier research resources to carry out your assignments. The work performed may potentially lead to a scientific publication.

Context

While generative AI has significantly advanced grasp synthesis in simulation, deploying these models onto physical hardware remains a critical bottleneck. End-to-end learning methods often require prohibitively large datasets of real-world demonstrations, making modular, perception-driven pipelines a highly promising alternative for complex manipulation. This internship bridges the gap between state-of-the-art generative grasp planners and real-world physical execution. The core objective is to develop a robust robotic skill layer — integrating 6D object pose estimation with dynamic hand control — that enables a bi-manual dexterous setup to reliably solve complex assembly tasks. The following works conducted at the lab will serve as a basis for the internship:

https://cea-list.github.io/cotograspweb/

https://cea-list.github.io/BOP-Distrib/

What do we expect fron you?

The internship is research oriented. The chosen candidate will work closely with a PhD candidate. The missions will be the following:

• Sim-to-Real Grasp Execution: Deploy and evaluate our existing generative AI grasp planners [1, 2] on physical dexterous hand hardware.

• Dynamic Object Perception: Integrate and adapt 6D pose estimation models (e.g., [3]) to accurately localize and track objects of interest.

• Robust Grasp Control: Design and implement a dedicated hand controller capable of executing dynamic and reliable grasps based on the generated plans.

• Bi-Manual System Integration: Scale and deploy the complete  perception-to-control pipeline onto a dual-arm setup to autonomously solve complex assembly tasks.

This internship is designed as a first step before a PhD.

Bibliography
[1] J. Mérand, B. Meden, L. Chen, and M. Grossard, “GOAG: Generative and object-agnostic grasp planner for dexterous robotic manipulation,” in International Conference on Intelligent Robots and Systems, 2026.

[2] J. Mérand, B. Meden, L. Chen, and M. Grossard, “CoToGrasp: Contact-topology-conditioned dexterous grasp synthesis via canonical workspace learning,” in European Conference on Computer Vision, 2026.

[3] A. Brazi, B. Meden, F. M. de Chamisso, S. Bourgeois, and V. Lepetit, “Corr2distrib: Making ambiguous correspondences an ally to predict reliable 6d pose distributions,” IEEE Robotics and Automation Letters, 2025.

#Cea List

Profile :

• Currently enrolled in your final year of a Master's program (M2) or an Engineering School (including gap year students).

• Practical experience in robotics, specifically with ROS2 knowledge.

• Strong skills in computer vision and machine learning, particularly with deep learning, perception models, and generative AI.

• High proficiency in Python alongside experience using a deep learning framework, with a strong preference for PyTorch.