World Models Research Engineer for Generative Robotics

Humanoid; United Kingdom; On-site; GBP 95,000 - 140,000; Full time

Benefits offered by this job

Stock options

30+ days off

Private healthcare

Pension scheme

Free breakfast and lunch

Frontier work environment

Direct access to leadership

Job summary

Humanoid is seeking a Research Engineer on the World Models team to build action-conditioned multimodal models that predict how the world evolves for our humanoid robots. You will design architectures, run large training jobs, and validate models against real fleet data from industrial deployments.

The role focuses on video, state, and language modeling, with collaboration across pretraining, RL, and manipulation teams to enable safer, scalable robotics systems.

Qualifications

• Track record of training large generative models with shipped artifacts or publications.
• Experience with modern generative architectures: diffusion, transformers, latent models, or video prediction.
• Experience with large-scale distributed training and handling big datasets.
• Strong Python + PyTorch/JAX; profiling and data loader optimization.
• Empirical rigor in evaluations, baselines, and clear experiment documentation.
• Interest in grounding generative models in physical reality, not just pixels.

Responsibilities

• Design and train multimodal world models — video, state, action, and language.
• Build action-conditioned video prediction and dynamics models with long-horizon consistency.
• Develop learned-simulator evaluations to predict real-world success before deployment.
• Generate synthetic rollouts and counterfactual data to augment policy training.
• Establish fidelity metrics and calibration protocols for world-model trust.
• Create data pipelines from telemetry, logs, and video into training corpora.
• Run scaling and ablation studies and communicate findings clearly.
• Collaborate with pretraining, RL, and manipulation teams to integrate models into training loops.

Skills

World models

Diffusion models

Autoregressive transformers

Python

PyTorch

Distributed training

Experiment design

Collaboration

Tools

JAX

MuJoCo

Isaac Sim

Diffusion

Job description

Humanoid is seeking a Research Engineer on the World Models team to build action-conditioned multimodal models that predict how the world evolves for our humanoid robots. You will design architectures, run large training jobs, and validate models against real fleet data from industrial deployments.

The role focuses on video, state, and language modeling, with collaboration across pretraining, RL, and manipulation teams to enable safer, scalable robotics systems.