As a member of the Autopilot Fleet Learning team, you will take on a highly cross-functional role to automatically label large amounts of data from the global Tesla vehicle fleet, train cutting-edge deep neural networks on these labels, and integrate the models with the rest of the Autopilot software stack to support numerous state-of-the-art autonomous driving functionalities and safety-critical features.
We are looking for both generalists with a breadth of expertise who are excited to work across the entire pipeline, as well as specialists who can dive deep on specific modules. An ideal candidate will possess strong expertise in at least one of the following areas.
Responsibilities
Develop offline state estimation, 3D reconstruction, and sensor fusion algorithms to automatically generate supervision for deep neural networks.
Train deep neural networks with large scale, auto-labelled datasets.
Design and implement tools, tests, and metrics to accelerate the data generation and model development cycles.
Integrate the models with the real-time embedded C++ software stack.
Work with planning & controls team to develop control policies on top of network outputs.
Requirements
Minimum 3 years of experience writing production-level Python or C++.
Strong mathematical fundamentals including linear algebra, vector calculus, probability theory, and numerical optimization.
Familiarity with basic computer vision concepts, such as camera intrinsics, extrinsics, projections, and epipolar geometry.
Exposure to a major deep learning framework such as PyTorch, TensorFlow, Keras, or MXNet.
Preferred Qualifications
Experience writing both production-level Python (including Numpy and Pytorch) and modern C++.
Proven track record of training and deploying real world neural networks.
Comfortable with general robotics, state estimation, filtering.
Prior work in Robotics, State estimation, Visual Odometry, SLAM, Structure from Motion, 3D Reconstruction.
Exposure to recent advances in Differentiable Rendering and Neural Rendering Techniques.