VLA Pretraining Lead: Shape Robotic Intelligence at Scale

Thehumanoid; Greater London; On-site; GBP 120,000 - 190,000; Full time

Benefits offered by this job

Equity options

Paid time off

Private healthcare

Pension scheme

Free daily meals

Job summary

Humanoid seeks a Pretraining VLA Lead to own the pretraining stage of our VLM/VLA framework powering a fleet of robots across platforms. Define base model strategy, data mixtures, and scaling, leading research engineers training foundation policies on diverse real-world and synthetic data.

You will design large-scale distributed training, establish evaluation suites, and collaborate with MLOps and Data Platform teams to ensure smooth transfer to post-training and real-robot inference.

Qualifications

• Track record of building deep-learning systems, with shipped models or published artifacts.
• Proven experience pretraining large models — LLMs, VLMs, video/generative models, or VLAs — at multi-node scale.
• Deep understanding of transformer and diffusion architectures, multimodal training, and distributed training practices.
• Strong Python + PyTorch/JAX; able to debug and profile ML systems and write maintainable research code.
• Data-driven scaling decisions and ability to communicate trade-offs to researchers and leadership.
• Document experiments clearly and build teams that do the same.

Responsibilities

• Own the VLA pretraining roadmap end-to-end: architecture choices, data mixtures, scaling laws, and evaluation protocols for base models.
• Push pretraining beyond a single recipe: explore transformer- and diffusion-based architectures, video pretraining, and world-model objectives for multimodal data.
• Lead, grow, and mentor a team of deep learning engineers focused on pretraining, setting research direction and engineering standards.
• Design and run large-scale distributed training on multi-node GPU clusters; drive throughput, stability, and cost efficiency with MLOps & Data Platform teams.
• Define pretraining-scale data requirements: partner with Data Collection and external providers for diverse trajectories.
• Build robust base-model evaluation suites predicting downstream real-robot performance and guide go/no-go scaling decisions.
• Establish continuous pretraining pipelines: dataset versioning, curation, deduplication, weak supervision labeling.

Skills

Deep learning systems

Pretraining large models

Transformer architectures

Diffusion architectures

Python

PyTorch/JAX

Data-driven decision making

Experiment documentation

Tools

PyTorch

JAX

Job description

Humanoid seeks a Pretraining VLA Lead to own the pretraining stage of our VLM/VLA framework powering a fleet of robots across platforms. Define base model strategy, data mixtures, and scaling, leading research engineers training foundation policies on diverse real-world and synthetic data.

You will design large-scale distributed training, establish evaluation suites, and collaborate with MLOps and Data Platform teams to ensure smooth transfer to post-training and real-robot inference.