MTS, Robot Learning (Paris)
- Hiring from
- France
- Work type
- Hybrid
- Posted
623,890 remote jobs, straight from company career pages
100% free · New jobs every minute
Show job descriptionHide job description
The role
You will teach robots new skills and make sure those skills work on real hardware.
Genesis is a physical AI lab building a foundation model for robotics: a "brain" that lets robots do real work with their hands. We build the whole stack in-house. That includes our model GENE-26.5 (released May 2026), our own open-source physics simulator, a dexterous robotic hand, a data glove built to the exact shape of that hand and Eno, our wheeled dexterous robot now going to its first customers.
As a Robot Learning Engineer, you turn human demonstrations and simulation into skills that run on real robots. You own the path from "here is a new task" to "the robot does it reliably".
What you'll do
Teach robots new skills. Train manipulation and locomotion policies with imitation learning and reinforcement learning, using glove demonstrations, teleoperation data and millions of simulated runs.
Make it work on the real robot. Deploy policies on our hand and on Eno, find where they break (latency, timing, perception, hardware) and fix it.
Close the sim-to-real gap. Work with our simulation team so the simulator and the real robot behave the same, then use that to train faster.
Build the teleoperation system that feeds the model. Smooth, precise, low-latency teleop: better control for operators means better data, and better data means better policies.
Own the learning-based control stack. Develop and optimize everything between the model's output and the motors.
Ship tasks end to end. Take a new task or customer demo from first idea to a robot that does it reliably, and turn what you learn into tools the whole team uses.
What your first six months could look like
Month 1: get a policy running on our hand, first in simulation, then on the real robot. Learn our loop from glove data to trained policy.
Months 2–3: own one manipulation task end to end. Measure its success rate on the real robot and push it up.
Months 4–6: improve a shared piece of the system the whole team relies on, such as teleop latency, sim-to-real tooling or evaluation. Or lead a task for one of Eno's first customers.
These are examples. Your projects will depend on your strengths and on where the robot needs help most.
What you'll bring
You've made learning work on a real robot. You trained a policy with reinforcement or imitation learning and ran it on physical hardware: an arm, a hand, a humanoid, a quadruped or a drone. You can name the robot, the task and how often it succeeded.
You build it yourself. Production-level Python or C++. You write the training code, the deployment code and the tools to debug both.
You debug across the whole system. When the robot fails, you can tell whether the cause is the model, the data, the latency, a sensor or the hardware.
Ownership and patience. You find problems before anyone asks, work methodically, and keep going until the result is reliable, not just a good demo video.
Some of our team have spent a decade shipping robot systems. Others joined after a master's where they made one robot do one hard thing really well. What we look for is proof that you make robots work.
Nice to have
Dexterous hands, tactile sensing or humanoid whole-body control
Learning from human demonstrations or human video, teleoperation systems, or vision-language-action (VLA) models
Sim-to-real transfer, differentiable simulation or large-scale RL in simulation
Training large models across many GPUs
First-author papers at CoRL, RSS, ICRA, IROS, T-RO or RA-L (helpful, not required)
Hardware you built yourself, or robotics competitions and hackathons you competed in
Hands-on troubleshooting across mechanical, electrical and software parts of a robot
Why Genesis
The whole loop under one roof. Glove, data, simulator, model, hand, robot, customer. Very few places let you touch every link.
Real robots, real use case. Our GENE-26.5 demo shows our hand doing lab work, solving a Rubik's Cube mid-air and cooking a multi-step meal.
Open source. Our simulator is public, so the tools you improve are used well beyond Genesis.
Practical details
Location: Paris (no full remote).
Type: full time, Engineering & Research
Don't tick every box? Apply anyway. We care more about what you've made robots do than about your title or years of experience.