Quentin Leboutet

I'm a research scientist at Intel in Munich, specializing in the generation of 3D and articulated assets using diffusion models.

At Intel I also contributed to the development of the SPEAR photorealistic simulator, the Open3D library and the OpenBot framework. I completed my PhD in Electrical Engineering and Computer Science at the Technical University of Munich, under the guidance of Prof. Gordon Cheng.

Email  /  CV  /  Scholar  /  Linkedin  /  Github  /  ResearchGate  /  HuggingFace

Research

I'm interested in generative AI, deep learning, agentic AI, and robotics. My PhD research focused on Robot Control, Tactile feedback, State Estimation and Inertial Parameters Identification.

MIDGArD is a modular diffusion framework that generates articulated 3D assets with enhanced controllability, enabling seamless integration into physics engines for advanced digital content and robotics applications.

IN-Sight is a self-supervised navigation system that supports interaction with obstacles using RGB-D data, enabling robots such as ANYmal to seamlessly navigate complex real-world environments.

OpenBot-Fleet is a scalable open-source cloud robotics system that leverages smartphones and affordable wheeled robots to learn and deploy robust navigation policies across real-world environments.

Introducing BIRDy: an open-source Matlab toolbox that benchmarks 17 cutting-edge inertial parameter identification methods, enabling precise robot dynamics analysis for both simulated and real-world manipulators.

A second-order kinematics estimation method that utilizes distributed inertial feedback and self-calibrating artificial skin to measure joint motions in humanoid robots, alleviating noise and lag issues.

An adaptive sensor-selection algorithm that dynamically optimizes inertial sensor usage on robots in real-time, enhancing scalability and robustness for high-order motion estimation in dynamic control applications.

A holistic approach to engineer the artificial skin for robots with an example of a multimodal skin cell showing multiple humanlike sensing modalities.

A quadratic programming-based tactile control framework that equips mobile robots with whole-body compliance through adaptive artificial skin, enabling robust and adjustable responses to multi-contact environmental interactions.

A quadratic programming-based framework that empowers omnidirectional mobile robots with whole-body compliance, utilizing artificial skin's tactile feedback to enable robust and prioritized interactions with complex environments.

Teaching Activities

Credits to Jon Barron for the website source code.