VLA Pre-training & World Models | Humanoid
Core
Pre-train base vision-language-action models and multimodal world models for humanoid robot manipulation pipelines, running on real robots.
Role type
Senior individual contributor deep learning engineer (robotics)
Builds
Base VLA models, action-conditioned world models, and learned simulators for real-world robot deployment
Domain
Robotics, Deep Learning, Multimodal AI
Deliverable
production ML models
Required skills
Deep learning system design, LLMs/VLMs/video generative models, distributed training infrastructure, Python, PyTorch or JAX, empirical rigor, proactive problem solving
Preferred skills
Robotics/autonomous driving experience, simulators (Isaac Sim, MuJoCo), RL for LLMs/robotics, VLA frameworks (OpenVLA, Physical Intelligence), on-device optimization, top-tier conference publications
Technologies
PyTorch, JAX, diffusion models, transformers, Isaac Sim, MuJoCo
Responsibilities
Pre-train models on multi-embodiment trajectory corpora and internet-scale video; design and train multimodal world models using diffusion and transformer architectures; build data pipelines for fleet telemetry and teleoperation logs; run scaling and ablation studies; optimize models for real-time edge inference on robots
Seniority
Senior, hands-on IC
