VLA Pre-training Lead (Deep Learning)
Lead the development and scaling of vision-language-action (VLA) foundation models for humanoid and wheeled robots using multimodal data from real-world trajectories, teleoperation, and synthetic sources. Define base model architecture, data strategies, and evaluation frameworks that enable downstream fine-tuning for robot capabilities. Work closely with MLOps, data, and reinforcement learning teams to build continuous pretraining pipelines and optimize for real-time inference on edge devices.