Staff Scientist: World Models for Robots & Driving

Odyssey; Greater London; On-site; GBP 90,000 - 150,000; Full time

Job summary

Odyssey is an AI lab pioneering generalized world models for robotics, driving, and perception. We seek researchers and engineers who will push diffusion world models, train policy-grounded agents, and deploy models to real robots and vehicles in a collaborative, high‑risk environment.

You will own end‑to‑end ML workflows, leverage state‑of‑the‑art diffusion backbones, and interface with robotics simulators and real hardware to test and refine policies at scale.

Qualifications

• PhD (or equivalent) plus 2+ years research or 4+ years software engineering with ML.
• Hands-on experience with model-based RL, learned simulators, diffusion or VLA policies.
• Comfort treating policy and model as one system.
• End-to-end ownership of projects.
• Sim-to-real experience is a strong plus.
• Willingness to work in-person in London, Zurich, or Bay Area.

Responsibilities

• Learn what makes interactive world models tick and how data and backbones interact.
• Advance world model usability as a robotics simulator with multi-sensor data.
• Implement diffusion, model-based RL, and related losses and tuning.
• Train policies inside the model and evaluate on real robots and in driving tests.
• Run ablations across world model and policy architectures and data mixes.
• Bridge to real robots and vehicles with data pipelines and policy deployment.
• Exploit GPUs to boost training and inference efficiency.
• Own the full ML stack used by Odyssey researchers and engineers.

Skills

Model-based RL

Diffusion models

Robotics

Policy learning

PyTorch

Experimentation

Education

PhD or equivalent research experience

Tools

PyTorch

JAX

Job description

Odyssey is an AI lab pioneering generalized world models for robotics, driving, and perception. We seek researchers and engineers who will push diffusion world models, train policy-grounded agents, and deploy models to real robots and vehicles in a collaborative, high‑risk environment.

You will own end‑to‑end ML workflows, leverage state‑of‑the‑art diffusion backbones, and interface with robotics simulators and real hardware to test and refine policies at scale.