Open to opportunities

Sébastien Combrexelle

@sbastiencombrexelle

Message

I build real-time 3D perception systems for robotics using computer vision and machine learning.

France

Message

What I'm looking for

I'm looking to build and deploy production-grade 3D perception and machine learning systems, especially where computer vision, synthetic data, simulation, and robotics enable measurable real-world impact.

At Squaremind, I own end-to-end R&D for a 3D perception module supporting robot trajectory planning in high-definition full-body scanning for dermatology. I developed and deployed real-time multi-view human mesh reconstruction from a six-camera RGB-D rig, robust to occlusions and flash lighting.

I built synthetic-data and training pipelines with Blender and an in-house OpenGL/moderngl simulator, generating over one million images with domain randomization. By optimizing model architecture, I reduced inference latency from 30ms to 12ms and reconstruction error from 4cm to 1.5cm, enabling safe robot operation within 5cm of patients.

Previously, I developed wind-production and anomaly-detection models at I4SEE, VAE-based visual anomaly detection at AnotherBrain, and statistical and CNN-based Lidar models at Leosphere. My PhD in statistics focused on hierarchical Bayesian models and MCMC inference, resulting in publications at ICLR, SIAM, and IEEE venues.

Experience

Work history, roles, and key accomplishments

Current

3D Machine Learning Engineer

Current

Squaremind

Jan 2023 - Present (3 years 7 months)

Owned end-to-end R&D of the 3D perception module for a medical robotics startup, developing multi-view human mesh reconstruction models and geometric calibration for a 6-camera RGB-D rig. Built synthetic data pipelines and optimized model architecture to reduce inference latency and reconstruction error.

3D Reconstruction Human Mesh Reconstruction RGB Calibration Extrinsic Calibration Hand Eye Calibration Blender OpenGL Synthetic Data Generation Domain Randomization PyTorch Pytorch Lightning Transformers PointNet++Onnx Python

Education

Degrees, certifications, and relevant coursework

Institut National Polytechnique de Toulouse

Doctor of Philosophy, Statistics

2013 - 2016

PhD in Statistics with a focus on multivariate multifractal analysis, developing hierarchical Bayesian models and MCMC methods. Published in top venues and received the Léopold Escande best PhD award.