Research Internships 2027 (W/M)

# Research Internships – Physical AI
(F/M/X)

Internship · Paris · 4–6 months · Flexible start in 2027

🔎 Who are we?

Kyutai is a non-profit research laboratory based in Paris, dedicated to open science in artificial intelligence. We work on high-impact projects at the frontier of AI and train the next generation of researchers, hosting interns, PhD students and postdocs.

We are the team behind open models such as Moshi, Hibiki, Pocket TTS or Helium, and more recently the vision-language-action model REKI and the world model MIRA. We are now growing a dedicated Physical AI team to push this direction further.

📝 Internship description:

As a research intern, you will join a team of experienced researchers and engineers. We will define your project together, based on your profile and the lab’s priorities, in areas such as:

- World models: learning predictive models of the physical world from video, interaction and sensor data.

- Vision-Language-Action (VLA) models: turning perception and language into robot actions.

- Motion retargeting: transferring human or animal motion to robots with different bodies.

- Reinforcement learning: training robust control policies in simulation and at scale. In every project, the aim is to take models from simulation to real robots.

What you will do:

- Define your subject: work with your supervisor to turn the initial direction into a precise research question, and have it approved by your school or university.

- Review prior work: study the relevant literature, understand existing methods and reproduce key results where useful.

- Form sharp hypotheses: critically analyze existing approaches, find where they fall short, and formulate clear, testable hypotheses with your supervisor.

- Build and train models: design architectures and training methods, build or curate the datasets you need (real, simulated or teleoperated), run experiments at scale and evaluate rigorously against baselines.

- Bring it to the robot: validate your work in simulation and, when it is ready, deploy it on our real robots. Closing the sim-to-real gap is part of the job.

- Share your work: where applicable, contribute to a research paper and release code and models on platforms such as GitHub or Hugging Face.

🎓 About you

- Background: Master's student (M2) or in the final year of an engineering school or university, specializing in machine learning, robotics, applied mathematics or computer science.

- Strong competence in at least one of:

Machine learning: deep learning, generative models, transformers, vision or language models.

- Robotics: control, kinematics and dynamics, motion planning, simulation.

- Robot learning: reinforcement learning, imitation learning, world models, VLAs, sim-to-real.

- Bridging these fields is a strong plus.

- Solid in math and algorithms: comfortable with the theory and able to turn it into working code.

- Hands-on engineer: fluent in Python and a deep learning framework (PyTorch or JAX). Experience with C++, simulators (MuJoCo, Isaac Sim or similar) or real robotic hardware is a plus.

- Fluent with AI agents: you build and work with AI agents to speed up both your coding and your research, and you know how to keep their output rigorous.

- Curious and rigorous: you care about why things work, hold yourself to high scientific standards, and enjoy working as part of a team.

🚀 What we offer you:

- Ambitious research with close supervision from experienced researchers in AI and robotics.

- Real robots: hands-on access to our robotic platforms to test and deploy your work in the physical world.

- Significant compute to run ambitious experiments.

- Open science: the possibility to publish your work, attend scientific conferences and release open-source code.

- A path forward: the option to continue with a PhD at Kyutai.

- Competitive internship allowance and benefits.

Interested?

We'd love to hear from you!