AI Engineer, Whole Body Controls, Optimus

Tesla AI is solving robust embodied intelligence through humanoid robots. On the Whole Body Controls team, you will build the models behind whole-body motion — the policies that let Optimus walk, balance, recover from disturbances, and manipulate objects — including control of the hands — on real hardware. Similar to the end-to-end models behind Full Self-Driving, we strive for simple, scalable algorithms; as an AI Engineer you will own the modeling side of that effort, designing network architectures, building and tuning training recipes, and shaping the data and representations that turn raw sensor streams into robust motion, working across the full training stack from pre-training to reinforcement learning. This role is for engineers with deep learning fundamentals; robotics domain knowledge is a bonus. We hire at all levels, including recent graduates with strong deep learning fundamentals and a demonstrated project or research record.

Responsibilities

Design and iterate on network architectures for whole-body motion policies, world models, and learned dynamics models Build, tune, and scale training recipes — losses, curricula, optimization, and pre-training-to-reinforcement-learning pipelines Build robust data pipelines and data engines that feed model training at scale Create evaluation benchmarks that enable fast, iterative model improvement Improve the robustness of learned policies, and deploy them on Optimus

Requirements

Proficiency in Python (numpy, pytorch) Deep learning fundamentals: architectures, optimization, training dynamics, and representation learning Proficiency with modern network architectures (such as transformers and sequence models), and a strong grasp of how to design and adapt them Experience training neural networks in PyTorch, including large-scale or distributed training Experience building robust data pipelines and data engines for large-scale model training Preferred: designing novel architectures or training recipes that shipped to production; world models, imitation learning, or reinforcement learning; any robotics, control, or multimodal-sensing exposure Equivalent experience through projects, research, or coursework is acceptable