- London
- On-site
- Mid-level
- Permanent
Owns the full training loop for vision-language-action policies, from pre-training a base model on multi-embodiment trajectory corpora through behaviour cloning and RL fine-tuning to deployment. Builds continuous data pipelines that ingest synthetic and teleop data, version and weakly-supervise it, and surface failure cases for retraining, working with MLOps and Data Collection teams on distributed training and edge inference.
PyTorch · JAX · OpenVLA +1