About

I am currently pursuing a PhD in AI for Science and Simulation at the Technical University of Madrid (UPM). I hold a Master’s degree (MSc) in Applied Mathematics and a Bachelor's degree (BSc) in Aerospace Engineering, both earned with top honors from UPM.

During my master's, I began working as an AI Researcher collaborating with Airbus R&D teams on projects involving reinforcement learning for robotics and autonomous aerial vehicles, as well as sequence modeling and surrogate modeling for physics-based problems.

My current research delves into generative modeling (particularly diffusion models), geometric deep learning, and reinforcement learning, with a focus on their application in scientific and phisics-based simulation contexts.

Publications

Giral, F., Manzano, A., Gómez, I., Koumoutsakos, P., & Le Clainche, S. (2025). Generative Urban Flow Modeling: From Geometry to Airflow with Graph Diffusion. arXiv preprint: arxiv.org/abs/2512.14725. (Under Review). [Preprint]
Giral, F., Gómez, I., & Clainche, S. L. (2025). Intercepting Unauthorized Aerial Robots in Controlled Airspace Using Reinforcement Learning. Aerospace Science and Technology, 110985. doi.org/10.1016/j.ast.2025.110985 [Paper] [Video]
Giral, F., Gomez, I., & Le Clainche, S. (2024). Control and Motion Planning of Fixed-Wing UAV through Reinforcement Learning. Results in Engineering, 102379. doi.org/10.1016/j.rineng.2024.102379 [Paper]
Giral, F., Gómez, I., Vinuesa, R., & Le Clainche, S. (2024). Transformer-Based Fault-Tolerant Control for Fixed-Wing UAVs Using Knowledge Distillation and In-Context Adaptation. arXiv preprint: arxiv.org/abs/2411.02975. (Under Review). [Preprint] [Video]

Projects

Academic projects

Bachelor's thesis: Deep Reinforcement Learning Tool Applied to Airline Revenue Management Optimization [Text]

Master's thesis: Automatic flight control systems based on deep reinforcement learning [Text]

Personal projects

DeepMind's GenCast reimplementation using JAX [Code]

Reimplementing GenCast diffusion model using the Flax NNX API.
Tools Used: JAX, Flax

gencast_gif

Training RL-based controllers directly on real hardware [Code]

Deployed Reinforcement Learning algorithms directly on real hardware to control the dynamics of a two-rotor system.
Tools Used: Python, Pytorch, C++

bicopter_gif

Autonomous racing using RL and Autoencoders [Code]

Implemented an autonomous racing car using a reinforcement learning controller and a pre-trained autoencoder to learn in a latent space and improve training sample efficiency.
Tools Used: Python, Pytorch, SB3, OpenCV

rl_racing