Senior Robotics Software Engineer

Make a humanoid robot know exactly where it is, and where its body is, in real time. Then watch your work train the AI that runs it.

This is a Senior or Staff perception and localization role on a humanoid robotics team: SLAM, visual-inertial odometry, state estimation and sensor fusion, on robots heading into real deployments.

• Fully remote within the US, with occasional travel to the team and test sites.
• Work on one of the hardest problems in robotics today, with humanoids moving into real deployments.
• Code that runs on the robot and also feeds the AI model that controls it.

• Real-time localization and state estimation. Architect and build onboard localization, ego-motion tracking and state estimation for a humanoid, fusing cameras, IMUs, joint encoders and other onboard sensors into a single estimate the rest of the robot can trust.
• Pose tracking for AI training data. Build and maintain scalable offline pose-tracking pipelines that process large volumes of training data from robots and other collection sources.
• Multi-sensor calibration. Develop and optimize calibration, including multi-camera and camera-to-IMU intrinsics and extrinsics, for both the robot and the platforms used to collect its data.
• Evaluation. Build data-driven evaluation and testing frameworks, so you can prove with numbers that localization is improving.
• Shipping to the real world. Write reliable, production-quality software that runs on robots doing real work.

• 6+ years building localization, SLAM or state estimation that ran on real hardware.
• Deep expertise in estimation and nonlinear optimization: factor graphs, bundle adjustment and filtering.
• Strong 3D geometry and visual-inertial SLAM, including feature matching and tracking and Structure from Motion.
• High-performance C++ for real-time systems.
• Hands-on sensor calibration experience, especially multi-camera and camera-to-IMU.
• The self-direction to do excellent work remotely, with clear written communication and comfort owning a problem end to end.

• Visual-inertial SLAM on legged robots, where foot impacts and dynamic motion make estimation much harder.
• An AR/VR or motion-capture background. Head and body pose estimation for headsets or mocap is very close to this problem.
• Learned methods in your toolkit: deep learning for SLAM, 3D reconstruction or human pose estimation, used alongside classical geometry.
• Experience with localization at fleet scale, including the edge cases that only appear once robots are deployed.

Senior and Staff engineers who have shipped localization on real hardware, including humanoid and legged robots, autonomous vehicles, drones, and AR/VR headset tracking.