Senior VLA Pre-training Engineer for a fast-growing humanoid robotics startup building robots for industrial environments (manufacturing, logistics, warehouses).
The role — responsibilities:
- Own the full lifecycle of foundation models for robotic behavior: pre-training and fine-tuning VLA models on large-scale multimodal datasets
- Build and maintain data pipelines: collection, curation, synthetic data generation
- Design distributed training infrastructure and manage checkpointing
- Sim-to-real transfer: validate policies in simulation before real-world deployment
- Collaborate with simulation engineers, data scientists, and embedded teams
- Ship models that power real robots in production
Requirements:
- 5+ years building deep learning systems with shipped models or published work
- Hands-on experience with LLMs, VLMs, or VLA models (architecture, training, fine-tuning)
- Large-scale distributed training: PyTorch DDP, data pipelines, checkpointing
- Experience with robotics simulators (MuJoCo, Isaac Sim, Gazebo) or real robotic systems
- Strong Python and PyTorch/JAX skills; able to debug numerics and write production code
- Self-driven, proactive, clear communicator
Nice to have: Robotics background, sim-to-real experience, OpenVLA or similar frameworks.
Conditions & benefits:
- Compensation and benefits below
- Location: London, 100% onsite (daily access to robots, labs, and prototypes)
- Relocation support and visa sponsorship
- Private healthcare, pension (8%), 23 days PTO, equity