Robot Learning in Simulation

Our approach

Another path to physical AI.

Built to work with large AI models trained by imitation, not against them: they can run inside our engine, and our skills can help train them.

Imitation learning Lexio

Learns from Human demonstrations A 3D model, in simulation

Hardware The robot used Any robot

Best for Open-ended tasks High-mix industrial tasks

How the technology works

Robot skills, learned in simulation.

Inputs

ObjectIts 3D model
Work cellMachines, fixtures
Robot & gripperAny brand or type
Task goalGrasp, place, insert…

Lexio skill engine · in simulation

Agentic orchestration

Modules Planning VLA Vision Other modules

Scene generation

Evolutionary search

Evaluation in simulation

Simulator-agnostic

Output

Validated skills
Sim-to-real filters & robustification
Real robotHardware-agnostic

Skill libraryeach skill speeds up the next

Same task, three set-ups

One task. Any robot, any gripper.

The same end-of-line task on a stamping line, solved for three set-ups: two robot brands, a two-finger gripper and a suction cup.

UR10e · Two-finger gripper

RB10 · Two-finger gripper

UR10e · Suction cup

Learns in simulation

No teleoperation data to collect for each task. Lexio complements imitation learning on structured, repeatable tasks.

Faster deployment

Integrators deploy new tasks faster. Operators switch to a new part reference without calling an integrator.

No vendor lock-in

One engine across robots, grippers and simulators. Get more from the robots you already have, or choose the one that fits the job.

Validated on real hardware

Research-grade, proven on real robots.

Skills found by our engine have been transferred from simulation to real robots, across very different hardware.

Skills: Vision-based grasping, placing, insertion, stacking
Hardware: Cobots, humanoids, 2- to 5-finger grippers including dexterous hands, suction grippers
Simulators: Genesis, NVIDIA Isaac Lab, Bullet
Scope today: Any task current simulators reproduce faithfully: rigid objects and modellable semi-rigid ones. Deformable or slippery objects and highly unstructured scenes come next.

Work with us

Two ways to work with Lexio.

01

Robotisation feasibility

Know which tasks to robotise, and with what hardware, before you invest.

Learn more

01 · Robotisation feasibility

A verdict for each task, built and tested in simulation with candidate robots and grippers.
Hardware and risks: recommended robot and gripper, expected time to deployment, barriers to remove.
A prioritised roadmap, with a simulation video of each feasible task.

Request a feasibility study

02

New skill deployment

A robot skill for each new part or task, on your hardware, with no programming.

Learn more

02 · New skill deployment

Already decided to robotise a specific use case? Lexio speeds up its deployment.

Talk about your use case

Use cases

One engine, many tasks.

The same engine at work on very different industrial tasks.

Metal fabrication

Sheet metal bending

Bending sequences and robot motions for each new part on a robotic press brake, without hours of programming.

Manufacturing, electronics

Quality inspection

The robot presents every face of each new part to the cameras, with no dedicated fixture per reference.

Machining

CNC loading and unloading

A new part reference handled without calling a robot programmer.

Metal production

End-of-line pick and place

Parts picked at the end of the line and placed in racks or containers, for every new reference.

Biotech R&D

Lab automation

A bench robot runs a new protocol after a short set-up, with no code.

Energy, electronics

Battery cell testing

Each cell placed in the tester, measured, then sorted pass or fail.

European deep tech

Built in Paris. Designed to keep industry in control.

Our engine is modular: each skill combines identifiable modules and is validated in simulation. This design is built for explainable physical AI, and to leave industrial groups in control of their data.

Our goal: to become the reference skill engine for industrial robots, worldwide.

Our story

Born from research at Sorbonne University.

Lexio Robotics began at ISIR, the robotics lab of Sorbonne University and CNRS, where our founders pioneered the use of evolutionary algorithms to teach robots how to move, adapt and learn. Four years of public research later, we are turning that work into an engine for industry.

Cofounder, CEO

Johann Huber

PhD in robot learning from Sorbonne University, with highest honours. He pioneered quality-diversity algorithms for robot manipulation and has been distinguished in several university eloquence contests.

Cofounder, CTO

François Hélénon

PhD from Arts et Métiers and AI engineering degree from CentraleSupélec. He developed advanced AI models for natural human–robot collaboration. Passionate about physical AI, he turns research into real-world impact.

Lexio Robotics wins the Industry 5.0 & Robotics prize.

More than 120 startups from around the world competed, with technologies chosen for their impact on automotive production. Lexio Robotics won the Industry 5.0 & Robotics category.

The prize was presented at the Mondial de l’Auto in Paris by Stellantis and PFA, who funded the competition and selected the winners in this category.

Photo: Louis Westerloppe