Our robots navigate hospitals, hotels, and retail floors — narrow corridors, glass walls, crowds, carts, and lighting that changes by the hour. You will own localization, mapping, and navigation inside Omakase OS: one engineer, full ownership, deployed robots. This role absorbs what was previously posted as two overlapping SLAM positions; we want one strong owner, not a SLAM department.
Why this role
- Full ownership of a stack that ships: your maps and planners run every day at customer sites, and deployment speed you build becomes company margin.
- You sit next to the manipulation and runtime teams — navigation is integrated, not siloed.
Responsibilities
- Own the localization/mapping stack end to end (we build on LiDAR-inertial odometry, e.g. FAST-LIO-class systems) for indoor human-shared environments
- Build robust state estimation: sensor fusion across LiDAR, cameras, IMU, and wheel odometry; degeneracy handling; relocalization
- Develop and tune path planning and obstacle avoidance for dynamic indoor spaces (people are not static obstacles)
- Build the mapping/calibration toolchain used at every customer-site deployment (fast site bring-up is a product feature)
- Optimize for edge compute (Jetson-class): CPU/GPU budgets shared with policy inference
- Validate in simulation and on robots; define navigation acceptance tests FDEs can run at deployment time
Requirements
- 3+ years hands-on SLAM / VIO / LiDAR-inertial odometry experience with real sensor data on real platforms (research or production; simulation-only does not qualify)
- Production-grade modern C++ and solid Python
- Strong foundations in 3D geometry, state estimation (EKF/UKF, factor graphs), and nonlinear optimization
- ROS / ROS 2 on Linux; you have debugged bad odometry at a real site before
- Point cloud processing (PCL / Open3D) and multi-sensor calibration experience
Nice to haves
While not specifically required, tell us if you have any of the following.
- FAST-LIO / LIO-SAM family internals; loop closure and map management at building scale
- Deployment on embedded GPUs (Jetson), CUDA optimization
- Navigation among dense pedestrians; social navigation literature awareness
- Docker, CI/CD; Japanese language is a plus for site visits
