3D Vision

Job description

Job responsibilities

• Design and build end-to-end 3D perception systems for real-world robotic environments.
• Develop geometric vision pipelines—SLAM, reconstruction, tracking—and integrate them with learned models.
• Implement and optimize deep learning models for depth, flow, correspondence, and 3D understanding.
• Prototype rapidly, measure performance in realistic conditions, and iterate to production-quality solutions.
• Collaborate closely with robotics, simulation, and platform teams to integrate 3D perception into the full stack.
• Move quickly and independently to solve hard 3D vision and geometry problems.

Job requirements

• Very strong coding skills in Python and/or C++.
• Experience with PyTorch and CUDA, including custom operators, optimized data pipelines, or model acceleration.
• Experience building 3D vision systems from scratch — not just using off-the-shelf libraries.
• Strong understanding of different sensor modalities (e.g., cameras, IMUs, wheel odometry, LiDAR, etc.), including their measurement models, calibration procedures, temporal synchronization, and noise characterization.
• Deep domain expertise in at least two of the following areas: Multi-view geometry, bundle adjustment, pose graph optimization, visual-inertial odometry.
• Dense 3D reconstruction, novel-view synthesis (NeRF, Gaussian Splatting, etc.).
• Training depth estimation, optical flow, or feature matching networks.