Deep Learning Engineer World Models

Skills

About the Role

You will design and train multimodal world models that predict future robot environments from observations and actions. You will build action-conditioned video and dynamics models, develop learned-simulator evaluation, generate synthetic rollouts, establish fidelity metrics, build training data pipelines, and run scaling studies.

Requirements

• Training large generative models
• Generative architectures including diffusion models, autoregressive transformers, latent-variable models, or video prediction
• Large-scale distributed training
• Streaming datasets
• Checkpointing and state management
• Python
• PyTorch or JAX
• Kernel profiling
• Data loader optimization
• Experiment evaluation and documentation

Responsibilities

• Design and train multimodal world models using diffusion and transformer architectures
• Build action-conditioned video prediction and dynamics models
• Develop learned-simulator evaluation for candidate policies
• Generate synthetic rollouts and counterfactual experience for policy training
• Establish fidelity metrics and calibration protocols
• Build training data pipelines from fleet telemetry, teleoperation logs, and video
• Run scaling and ablation studies
• Integrate world models into policy training and evaluation loops

Benefits

• Stock options
• 30+ paid days off including 23 days of annual leave, UK bank holidays, and company closure days
• Private healthcare
• Pension scheme with 8% total contribution
• Free daily breakfast, catered lunch, and snacks in-office