About

Welcome to my personal website!

My name is Lucas Amoudruz, I'm currently a research associate at Harvard University in the CSE laboratory. My research interests include numerical simulations of microfluidics, red blood cells modeling, blood flow simulations, artificial microswimmers, reinforcement learning, high performance computing, cloud computing and Bayesian inference. The most fun part for me is to combine multiple aspects of the above!

Email: amlucas@seas.harvard.edu
Scholar: Lucas Amoudruz
Github: amlucas
Full CV (PDF): Last updated in July 2025

Research

I develop computational methods and software to simulate and control microscale systems, including blood flow and artificial microswimmers. My work combines high-performance computing, fluid dynamics, Bayesian inference, and artificial intelligence to enable biomedical applications such as targeted drug delivery and flow manipulation in microfluidics.

High performance computing for simulating blood...

Simulating blood at the cellular resolution is a complex task, as it involves resolving hydrodynamics between deforming cells. Blood is mainly composed of red blood cells (RBCs), and one of the contributions during my PhD was to accurately model these cells and their interactions at large scale. We use particles to represent RBCs and all fluids, via the dissipative particle dynamics (DPD) method. The membranes are discretized into a triangle mesh and elastic forces arise from bending and stretching energy potentials. Simulations typically consist of millions to billions of these particles, and we thus implemented a custom software package, Mirheo. Mirheo is implemented in C++/CUDA/MPI and achieves above 98% weak scaling efficiency across 1000 GPUs. Check out our 2020 paper in Computer Physics Communications for benchmarks and implementation details. Recently we have been tested the performance of Mirheo on the public cloud (article currently under review).

Mirheo has been used in a number of applications, and I am excited to make new predictions of blood flow, design novel microfluidics devices, and understand the biomechanics of RBCs. For example, we have used Mirheo to partly answer the question: Are RBCs optimal for transporting oxygen? Numerical simulations suggest that their shape, and in particular their area-to-volume ratio, maximizes oxygen transport in geometries similar to arterioles (more details in our 2024 publication in Biophysical Journal). Similarly, in our 2025 publication in Physical Review Fluids, Mirheo was used to study the effect of objects rotation in inertial focusing microfluidics.

...and artificial microswimmers

Artificial bacterial flagella (ABFs) are micron-sized devices equiped with a magnetic head and a corkscrew shaped body, which makes them able to propel when they are immersed in a rotating magnetic field. They are great candidates to reach specific regions in the body in a non-invasive way by swimming through the circulatory system of the body. Applications are numerous: targeted drug delivery, microsurgery, remote sensing, cells manipulation... I have extended our custom software Mirheo to simulate ABFs in blood. Here is a video of an ABF swimming in blood and another one of many ABFs in a bifurcation. One particular focus of my research tries to answer the following question: how to control the external magnetic field to guide ABFs to a target?

One possible approach is to use reinforcement learning (RL). RL learns a control policy by interacting with an environment. Typically, RL converges after thousands of episodes, while one simulation of ABFs in blood require thousands of GPU-hours. Instead we train the RL agent on reduced order models, which are much cheaper. We then test the learned policy in large scale simulations, as shown on the picture on the right. Details are available in our 2025 article in Physics of Fluids. Check out the video of the policy in action. RL was also helpful to design policies that control multiple ABFs independently, a difficult task given that the magnetic field is essentially uniform at these scales. Additional details can be found in our 2022 article in Advanced Intelligence Systems, where we have exploited the swimming properties of ABFs of different shapes.

ODIL: a method for control and path planning

Reinforcement learning is a great tool for control, but does not always converge in the settings I am interesting in. Furthermore, I typically have a model of the systems I want to control. This led us to developping a novel method that learns policies from a path planning objective and from the dynamics of the system in the form of an ODE or a PDE.

The method relies on ODIL, developped at CSE-Lab, originally used to solve PDE-based inverse problems. In the context of control and path planning, we use ODIL to solve both the system dynamics and the control policy at the same time. This approach is more efficient than RL in numerous benchmarks, summarized in our 2025 article published in Physical Review Letters.

To check the performance of that method, we have built a robotic device (shown in the right picture) that creates flow with rotating disks. Using the policy found by ODIL, the device can transport small beads to precise targets, as explained in our 2025 publication in Journal of Fluid Mechanics. I have built this robot with Petr Karnakov in less than two months to participate at the MassRobotics Form and Function Challenge where we have placed second over more than 40 participants.

Hierarchical Bayesian inference for red blood cells models

Calibrating the red blood cell (RBC) membrane model is crucial to predict blood flows. Prior work has been fitting the membrane parameters to experimental datasets, but we found that this approach leads to a poor transferability of the model. Instead, we have combined multiple experimental datasets to find the model parameters. Furthermore, the model and the data contain numerical, modeling and measurement errors.

To handle these sources of uncertainties, and the multiple datasets, we have taken a hierarchical Bayesian approach, represented on the left diagram. This structure models the variability of the model parameters within populations of RBCs. We found that hierarchical Bayesian inference gave a transferable model which can predict RBC dynamics in situations that were not used during the calibration of the model. More details are available in our 2021 publications in Physical Review Applied and our 2023 publication in Biophysical Journal.

Publications

2025

@article{amoudruz2025b,
  author = {Amoudruz, Lucas and Litvinov, Sergey and Koumoutsakos, Petros},
  doi = {https://doi.org/10.1063/5.0274623},
  eprint = {https://doi.org/10.48550/arXiv.2404.02171},
  journal = {Physics of Fluids},
  number = {7},
  publisher = {AIP Publishing},
  title = {Optimal navigation of magnetic artificial microswimmers in blood capillaries with deep reinforcement learning},
  volume = {37},
  year = {2025}
}

@article{amoudruz2025a,
  author = {Amoudruz, Lucas and Karnakov, Petr and Koumoutsakos, Petros},
  doi = {https://doi.org/10.1017/jfm.2025.10174},
  eprint = {https://doi.org/10.48550/arXiv.2506.15958},
  journal = {Journal of Fluid Mechanics},
  pages = {A15},
  publisher = {Cambridge University Press},
  title = {Contactless precision steering of particles in a fluid inside a cube with rotating walls},
  volume = {1014},
  year = {2025}
}

@article{alexeev2025a,
  author = {Alexeev, Dmitry and Litvinov, Sergey and Economides, Athena and Amoudruz, Lucas and Toner, Mehmet and Koumoutsakos, Petros},
  doi = {https://doi.org/10.1103/PhysRevFluids.10.054202},
  eprint = {https://doi.org/10.48550/arXiv.2408.09552},
  journal = {Physical Review Fluids},
  number = {5},
  pages = {054202},
  publisher = {APS},
  title = {Inertial focusing of spherical particles: The effects of rotational motion},
  volume = {10},
  year = {2025}
}

@article{karnakov2025a,
  author = {Karnakov, Petr and Amoudruz, Lucas and Koumoutsakos, Petros},
  doi = {https://doi.org/10.1103/PhysRevLett.134.044001},
  eprint = {https://doi.org/10.48550/arXiv.2506.15902},
  journal = {Physical Review Letters},
  number = {4},
  pages = {044001},
  publisher = {APS},
  title = {Optimal navigation in microfluidics via the optimization of a discrete loss},
  volume = {134},
  year = {2025}
}

2024

@article{amoudruz2024a,
  author = {Amoudruz, Lucas and Economides, Athena and Koumoutsakos, Petros},
  doi = {https://doi.org/10.1016/j.bpj.2024.04.015},
  eprint = {https://doi.org/10.48550/arXiv.2305.02197},
  journal = {Biophysical journal},
  number = {10},
  pages = {1289--1296},
  publisher = {Elsevier},
  title = {The volume of healthy red blood cells is optimal for advective oxygen transport in arterioles},
  volume = {123},
  year = {2024}
}

2023

@article{amoudruz2023a,
  author = {Amoudruz, Lucas and Economides, Athena and Arampatzis, Georgios and Koumoutsakos, Petros},
  doi = {https://doi.org/10.1016/j.bpj.2023.03.019},
  eprint = {https://doi.org/10.48550/arXiv.2303.03404},
  journal = {Biophysical journal},
  number = {8},
  pages = {1517--1525},
  publisher = {Elsevier},
  title = {The stress-free state of human erythrocytes: Data-driven inference of a transferable RBC model},
  volume = {122},
  year = {2023}
}

2022

@phdthesis{amoudruz2022b,
  author = {Amoudruz, Lucas},
  doi = {https://doi.org/10.3929/ethz-b-000550202},
  school = {ETH Zurich},
  title = {Simulations and control of artificial microswimmers in blood},
  year = {2022}
}

@article{amoudruz2022a,
  author = {Amoudruz, Lucas and Koumoutsakos, Petros},
  doi = {https://doi.org/10.1002/aisy.202100183},
  eprint = {https://doi.org/10.48550/arXiv.2101.10628},
  journal = {Advanced Intelligent Systems},
  number = {3},
  pages = {2100183},
  publisher = {Wiley Online Library},
  title = {Independent control and path planning of microswimmers with a uniform magnetic field},
  volume = {4},
  year = {2022}
}

2021

@article{economides2021a,
  author = {Economides, Athena and Arampatzis, Georgios and Alexeev, Dmitry and Litvinov, Sergey and Amoudruz, Lucas and Kulakova, Lina and Papadimitriou, Costas and Koumoutsakos, Petros},
  doi = {https://doi.org/10.1103/PhysRevApplied.15.034062},
  journal = {Physical Review Applied},
  number = {3},
  pages = {034062},
  publisher = {APS},
  title = {Hierarchical Bayesian uncertainty quantification for a model of the red blood cell},
  volume = {15},
  year = {2021}
}

2020

@article{alexeev2020a,
  author = {Alexeev, Dmitry and Amoudruz, Lucas and Litvinov, Sergey and Koumoutsakos, Petros},
  doi = {https://doi.org/10.1016/j.cpc.2020.107298},
  eprint = {https://doi.org/10.48550/arXiv.1911.04712},
  journal = {Computer Physics Communications},
  pages = {107298},
  publisher = {Elsevier},
  title = {Mirheo: High-performance mesoscale simulations for microfluidics},
  volume = {254},
  year = {2020}
}

@inproceedings{walchli2020a,
  author = {W{\"a}lchli, Daniel and Martin, Sergio M and Economides, Athena and Amoudruz, Lucas and Arampatzis, George and Bian, Xin and Koumoutsakos, Petros},
  booktitle = {Proceedings of the Platform for Advanced Scientific Computing Conference},
  doi = {https://doi.org/10.1145/3394277.3401849},
  pages = {1--12},
  title = {Load balancing in large scale bayesian inference},
  year = {2020}
}

2017

@inproceedings{economides2017a,
  author = {Economides, Athena and Amoudruz, Lucas and Litvinov, Sergey and Alexeev, Dmitry and Nizzero, Sara and Hadjidoukas, Panagiotis E and Rossinelli, Diego and Koumoutsakos, Petros},
  booktitle = {Proceedings of the Platform for Advanced Scientific Computing Conference},
  doi = {https://doi.org/10.1145/3093172.3093226},
  pages = {1--13},
  title = {Towards the Virtual Rheometer: High Performance Computing for the Red Blood Cell Microstructure},
  year = {2017}
}