About me
I am currently an associate researcher within the University of Edinburgh and Heriot-Watt University .
My area of research includes Legged locomotion (humanoids and quadrupeds), Trajectory Optimisation and Machine learning.During my PhD, I worked across the entire locomotion pipeline, from vision-based perception (extracting convex surfaces from heightmaps), to trajectory planning using optimal control, and finally to low-level control through ROS Control for full system integration. I primarily used Python for prototyping and C++ for performance-critical integration. This work has been succefully transfered to the industry on humanoids by PAL robotics.
In the context of machine learning, I am experienced with JAX and PyTorch. For motion planning and control, I have worked extensively with Crocoddyl, MuJoCo, Pybullet and Pinocchio.
Interfacing Whole-Body MPC with a Mixed-Integer SL1M for surface selection for the European h2020-Memmo project. Full integration with real-time perceptive pipeline for vision. Code available here.
Real time contact planning and control of the Solo robot in 3d environments. Reduced model MPC, interacting with a mixed-integer program.
Combining Trajectory Optimisation and Reinforcement Learning. An efficient MPC is guided by high-level RL commands to reach the green ball, with heightmap information. Code available here.
The University of Edinburgh/ Heriot Watt University — July 2024 to Present
Collaboration with Dr. Steve Tonneau. Integration of a differentiable simulator into policy-learning frameworks, enabling backpropagation of gradients through simulation to improve sample efficiency compared to traditional zero-order methods for Locomotion and Manipulation. Most of the development was carried out in JAX and MuJoCo-MJX framework.The University of Edinburgh — 2021 to 2024
PhD under the supervision of Dr. Steve Tonneau and Dr. Thomas Flayols within the Institute of Perception, Action and Behaviour (IPAB). Real-time synthesis of legged locomotion maneuvers in complex, unstructured en- vironments. Developed and integrated a fully autonomous locomotion pipeline including perception, motion and contact planning, and reactive control.
LAAS-CNRS (Toulouse, France) — 2020
Master’s thesis under the supervision of Nicolas Mansard. Implementation of a Model Based Predictive Controller for a quadruped robot using a reduced model dynamics.
Toulouse, France — 2016 to 2020
Selective three year program leading to a French master’s degree, with a multidisciplinary curriculum in Computer Science, Applied Mathematics, and Physics