Research Scientist / Engineer - Large-scale Training
Core
Develop architectures and training methods to make robotic foundation models more capable and efficient to train, focusing on pretraining, mid-training, and post-training for humanoid robots.
Role type
Senior IC research scientist/engineer (large-scale model training)
Builds
Robotic foundation models optimized for compute efficiency and real-robot performance
Domain
Robotics + Large-scale Deep Learning
Deliverable
production ML models
Required skills
Large-scale foundation model training, deep learning research, Python, PyTorch, model architecture design, training optimization, hardware-aware implementation, experimental design, scaling laws analysis, training dynamics diagnosis
Preferred skills
Triton, CUDA, fused kernels, efficient attention mechanisms, mixture-of-experts, distributed training, low-precision training, vision-language-action models, open-source contributions
Technologies
PyTorch, Triton, CUDA
Responsibilities
Research efficient model architectures for robotic foundation models; Optimize training recipes and convergence; Co-design architectures and compute kernels; Investigate efficient attention and sequence processing; Develop scaling laws and compute-allocation studies; Research training acceleration techniques; Run controlled ablations and performance benchmarks; Diagnose training dynamics and numerical instabilities; Evaluate transfer of pretraining improvements to real-robot success; Integrate validated methods into scalable training systems
Seniority
Senior, hands-on IC
