Research Engineer - Embodied World Models

UMA; Paris; Sur place; EUR 70 000 - 110 000; Plein temps

Résumé du poste

UMA is advancing physical AI by teaching robots to understand, plan, and reason in the physical world. The World Models team turns bold research ideas into working systems and real-world humanoid robots.

As a Research Engineer, you will build the pipelines, baselines, and experiments that move hypotheses to results, both on real robots and in simulation, shaping the future of robot intelligence.

Qualifications

• Strong software engineering and ML fundamentals; fluent in Python and PyTorch.
• Solid grasp of self-supervised representation learning; exposure to video models, RL, planning or robotics is a plus.
• MSc or PhD in ML, CS, robotics; what you built and shipped matters most.
• Experience with world models (e.g., JEPA) is highly valued.
• Curiosity, rigor, and a bias for getting things working.

Responsabilités

• Implement and reproduce state-of-the-art world-model baselines, run ablations and deploy them on real robots.
• Build and own the training, data, and evaluation pipelines for self-supervised learning on video and action data.
• Bring latent-space planning and model-predictive control experiments to life on real hardware and in simulation.
• Profile, debug, and scale distributed training runs; keep every experiment clean, logged, and reproducible.

Connaissances

Python

PyTorch

Distributed training

GPU training

Self-supervised learning

Video models

Reinforcement Learning

Robotics

Formation

MSc or PhD in ML, CS, robotics

Outils

JEPA

Description du poste

Your Mission

At UMA, we’re pushing the frontier of physical AI by teaching robots to understand, plan, and reason in the physical world. A critical part of this effort is building world models that can predict the outcomes of actions before they are executed on a robot.

About the Role

As a Research Engineer on the World Models team, you’ll be the force that quickly turns bold research ideas into working systems. You’ll build the pipelines, baselines and experiments that let our scientists go from hypothesis to results and humanoid robots that leave the lab for the real world.

Key Responsibilities

• Implement and reproduce state‑of‑the‑art world‑model baselines, run ablations and deploy them on real robots.
• Build and own the training, data, and evaluation pipelines for self‑supervised learning on video and action data.
• Bring latent‑space planning and model‑predictive control experiments to life on real hardware and in simulation.
• Profile, debug, and scale distributed training runs; keep every experiment clean, logged, and reproducible.

What You Bring To The Table

• Strong software engineering and machine learning fundamentals, fluent in Python and PyTorch, comfortable with distributed/GPU training.
• A solid grasp of self‑supervised representation learning; exposure to video models, RL, planning or robotics is a real plus.
• MSc or PhD in ML, CS, robotics or a related field—what matters most is what you’ve built and shipped.
• Experience with world models (e.g., JEPA) is highly valued.
• Curiosity, rigor, and a bias for getting things working.
• Bonus: You’ve worked in early‑stage startups or the core teams of big companies building humanoids or highly‑dexterous robots.
• UMA is an inclusive workplace that values exceptional builders over perfect pedigrees. Whatever your background, identity, or journey, if you don’t meet every criterion but believe you can have an outsized impact here, we strongly encourage you to apply.