VLA Pre-training & World Models | Humanoid

The role

This senior individual contributor role involves building base vision-language-action models or multimodal world models. You will train models from scratch on multi-embodiment trajectory corpora and internet-scale video to run on real robots.

What you will do

• Pre-train base models on multi-embodiment corpora and fine-tune policies
• Post-train policies with behaviour cloning and RL
• Partner with the Data Collection team to specify data and identify failure modes
• Run pre-, mid- and post-training on the VLA stack
• Build and maintain continuous pipelines for synthetic data and teleop logs
• Work with MLOps and Data Platform teams to scale distributed training
• Design and train multimodal, action-conditioned world models
• Build video prediction and dynamics models
• Develop learned-simulator evaluation to score candidate policies offline
• Generate synthetic rollouts and counterfactual experience

Required qualifications

• 3+ years building deep learning systems in industry or research
• Deep hands-on experience with LLMs, VLMs, image or video generative models
• Experience training models including pre-training runs and distributed infrastructure
• Experience with deep learning infrastructure including streaming datasets and distributed training
• Strong Python and PyTorch or JAX
• Strong coding fundamentals
• Empirical rigor
• Proactive approach to open problems
• Working English

Nice to have

• Robotics or autonomous driving experience
• Familiarity with simulators (Isaac Sim, MuJoCo) and sim-to-real
• Experience applying RL to LLMs or robotics
• Experience with VLA models; familiarity with OpenVLA, Physical Intelligence (π) models or similar
• Experience with world models for robotics or autonomous driving
• Proven productization of deep nets
• Publications at top-tier DL conferences (NeurIPS, ICML, ICLR, CoRL, CVPR) or equivalent open-source contributions

Benefits

• Stock options
• 30+ paid days off
• 23 days annual leave
• UK bank holidays
• Company shutdown between Christmas and New Year
• Private healthcare
• Pension: 8% total contribution (5% employee, 3% employer)
• Free breakfast, catered lunch and snacks in the office

Before you apply

• The process includes a LeetCode-style coding round
• Visa sponsorship and relocation to London

About the company

Humanoid is building the world’s most capable, commercially scalable and safe humanoid robots. Its platform, HMND-01, is already deployed in real industrial environments.