AI Engineer, Vision & Foundation Models, Optimus

Tesla is solving robust embodied intelligence through humanoid robots. You will join the team building the vision and multimodal foundation models that allow Optimus to understand, interact, and navigate the world. Working at the intersection of large-scale deep learning and real-world robotics, you will own the full model lifecycle - from designing massive-scale data engines and auto labeling pipelines to architecting novel neural networks and deploying them to thousands of robots.

Responsibilities

Design and build scalable data pipelines, utilizing classical computer vision and off-the-shelf models to auto label massive video datasets Develop and train state-of-the-art foundational vision and multimodal networks, focusing on efficient tokenization, compression, and fusion of vision, language, audio, and tactile data Iterate on neural network architectures to improve perception, 3D understanding, and world modeling capabilities Optimize models for deployment on edge compute, utilizing quantization, pruning, and distillation to ensure real-time performance on the robot Collaborate with hardware and controls engineers to close the loop between perception and actuation in the real world

Requirements

Strong software engineering skills in Python with an ability to write production-quality code Experience with PyTorch or another major deep learning framework (e.g., JAX, TensorFlow) Experience training large-scale models on distributed GPU clusters Strong mathematical fundamentals, including linear algebra, computational geometry, probability theory, and numerical optimization, are a plus Familiarity with classical computer vision (camera models, calibration, sensor fusion) is a plus Strong mathematical fundamentals, including linear algebra, computational geometry, probability theory, and numerical optimization, are a plus Familiarity with classical computer vision (camera models, calibration, sensor fusion) is a plus Domain expertise in one or more of the following areas: Foundational Vision: Vision encoder pretraining, dense prediction, segmentation 3D & Geometry: Multiview geometry, NeRFs, 3D Gaussian Splatting, geometric vision models Multimodal Models: World models, video generation, vision-language-action models Human/Object Reconstruction: Mesh reconstruction, hand pose estimation, human-object interaction, object affordance