Foundation is seeking an AI Engineer to build and maintain the foundational SLAM system for a humanoid robot operating in unstructured, real-world environments. You will focus on creating custom, embodied SLAM solutions from the ground up to enable centimeter-level precision localization and mapping in dynamic settings.
What You'll Do
- Build and maintain accurate, real-time maps of dynamic environments for a humanoid robot.
- Simultaneously localize the robot with centimeter-level precision.
- Develop custom, embodied SLAM solutions from the ground up for deployment across varied terrains like floors, stairs, cluttered rooms, and outdoors.
- Scale the SLAM infrastructure layer to be production-ready to support upstream systems like perception, planning, and manipulation.
What We're Looking For
- Deep knowledge in vision for robotic systems.
- Hands-on experience implementing SLAM pipelines in C++ and Python end-to-end.
- Strong working knowledge of modern SLAM frameworks (e.g., ORB-SLAM3, RTAB-Map, Cartographer, LIO-SAM, KISS-ICP) and ability to extend or rewrite core components.
- Comfortable with probabilistic state estimation, Kalman filtering (EKF/UKF), and particle filters for real-time localization.
- Familiar with loop closure detection methods and place recognition networks for long-term map consistency.
- Hands-on experience with simulation environments (e.g., Isaac Lab, MuJoCo) for development, testing, and sim-to-real validation.
Nice to Have
- Experience with neural or learned SLAM approaches (e.g., DROID-SLAM, iMAP, NeRF-SLAM).
- Experience with legged or humanoid-specific odometry challenges.
- Experience with multi-session and multi-agent mapping.
Technical Stack
- C++, Python
- SLAM Frameworks: ORB-SLAM3, RTAB-Map, Cartographer, LIO-SAM, KISS-ICP, DROID-SLAM, iMAP, NeRF-SLAM
- Simulation: Isaac Lab, MuJoCo
