About me
Biography
Julien Baumeyer works at the meeting point of science and industry. He holds a PhD in robotics and practises applied research: what he enjoys most is the moment a lab idea becomes a machine that runs on a factory floor.
His approach starts from first principles. Before choosing an algorithm, he goes back to the physics of the problem: the dynamics and inertia of a robot arm, the friction and backlash of its gears, the leakage of a magnetic field around a crack, the way light, optics and sensors turn a surface into an image. Modelling the whole chain, from the physical phenomenon to the measured signal, shows where the real difficulty lies and which tools can actually solve it.
The same goes for artificial intelligence, which he approaches through its mathematics rather than as a black box. Linear algebra, optimisation, probability and statistics explain why a network learns, why it fails and how far its predictions can be trusted. That foundation guides how he builds datasets, chooses architectures and loss functions, tunes training and measures uncertainty, so that AI becomes an engineering tool whose behaviour can be explained and validated.
That understanding is what lets research cross over into industry. A method that works in a paper also has to cope with noisy signals, cycle times, operator safety, standards and cost. He brings the scientific rigour of the laboratory (modelling, experimentation, validation and publication) to these constraints, and builds solutions that are both well-founded and deployable.
Today this work continues within LabCom OPTEND, a joint laboratory funded by the French National Research Agency (ANR) that brings together the ImViA laboratory, CMPhy, a maker of NDT equipment, and the ELyTMaX laboratory (INSA Lyon). The aim is to make non-destructive testing faster, more reliable and better quantified by combining magnetic particle, penetrant and eddy-current inspection with AI. Since 2026 he has been the technical supervisor of two PhD theses there: one on next-best-view planning to automate robotic inspection intelligently, the other on estimating the mechanical properties of materials from magnetic measurements.
He also passes this on in the classroom: since 2024 he has taught image processing at Université Bourgogne Europe and leads the Python and AI-projects courses of the English-taught Health and Artificial Intelligence master’s programme.
His fields are computer vision, robotics and artificial intelligence. He trains neural networks to spot defects a hurried eye would miss, fuses data from cameras and depth sensors, and designs control laws so that robots weighing hundreds of kilos can be guided by hand.
He is also a software developer, at ease anywhere from embedded firmware to Linux servers, and from prototype to product.
On the entrepreneurial side, he was CTO of Nexomedis and took part in the Smarche adventure, a start-up building sensors for gait rehabilitation. He came away with a taste for small teams and fast decisions.
As an R&D project manager, he bridges researchers, engineers and customers, and turns technical questions into plans.
He is an open-source advocate who self-hosts his own tools and favours open, well-understood solutions.
Away from the screen, he runs trails and flies paragliders. Trail running teaches endurance and pacing; paragliding, how to read conditions and make calm decisions. He brings both qualities to every project.
His conviction: small and medium-sized businesses deserve access to the same advanced technology as large groups, as long as it is pragmatic, open and useful.