3.23 Robotics State Estimation & Localization Engineer

Field AI•Irvine, CA

50d•$70,000 - $300,000•Hybrid

About The Position

We’re building the estimation and navigation stack that keeps our legged and humanoid robots balanced, aware, and mission-ready—indoors and out, with or without GPS. You’ll design and ship real-time estimators and fusion pipelines that combine IMU and GNSS/GPS/RTK with legged-robot proprioception (joint encoders, torque/force & foot-contact sensors) and exteroception (cameras, LiDAR , radar/ UWB ). You’ll take algorithms from log-replay to rugged field performance on embedded/Linux targets, partnering closely with controls, perception, and planning.

Requirements

Strong fundamentals in estimation & sensor fusion ( EKF/UKF , error-state, observability/consistency, covariance tuning).
Hands-on with IMUs (strapdown mechanization, bias/scale, coning/sculling) and GNSS/GPS/RTK (loosely vs tightly coupled INS ).
Experience with legged-robot proprioception: joint encoders, foot contact/pressure, torque/force sensors; using kinematic/dynamic constraints in estimators.
Proficiency in modern C++ (14/17/20) on Linux ; Python for tooling, analysis, and log processing.
Comfort with SO(3)/SE(3) , Lie-group math, and non-linear optimization.
Integration with at least two of: cameras ( VIO ), LiDAR ( LIO /scan-matching), UWB , magnetometer/barometer, radar.
Familiarity with ROS 1/ROS 2 , CMake/Bazel, Docker, CI/CD , and reproducible experiments.
Proven track record shipping research-to-production algorithms on real robots with field test cycles.
BS/MS/PhD in Robotics/EE/CS/AE or equivalent practical experience.

Nice To Haves

Factor-graph SLAM/VIO (GTSAM/iSAM2) and non-linear solvers (Ceres/g2o); hybrid filtering + smoothing in production.
Whole-body/legged tooling, momentum/ COM filters, terrain estimation, and contact-rich datasets.
Robustness techniques: adaptive noise models, M-estimators/gating, data association, map-based relocalization.
Experience with GPS-denied navigation at scale (warehouses, construction, urban canyons).
Real-time/performance chops ( RT-PREEMPT , lock-free pipelines, deterministic logging, on-robot telemetry).
Embedded/GPU acceleration ( Nvidia / CUDA ) for perception-aided estimation.
Designing calibration & end-of-line test procedures for production.

Responsibilities

State Estimation for Legged/Humanoid Bases
Design and tune EKF/UKF error-state filters for floating-base pose/velocity, COM , IMU biases, and contact states.
Fuse IMU, joint encoders, foot F/T & contact sensors; implement ZUPT/ZARU , slip handling, and kinematic/dynamic constraints.
Expose clean interfaces (frames/timestamps/covariances) to whole-body control and footstep planning.
Perception-Aided Localization & Mapping
Stand up VIO/LIO pipelines (stereo/ RGB-D + LiDAR) for GPS-denied operation, with map-based relocalization and loop closure.
Add global aids— GNSS/RTK , UWB beacons, prior maps—and blend filtering with factor-graph smoothing when advantageous.
Manage drift/consistency with robust outlier rejection, gating, and integrity monitoring.
Calibration, Timing & Robustness
Own time sync ( PTP /Chrony/hardware triggers) and multi-sensor calibration (Allan variance for IMU, camera-IMU/LiDAR-IMU/base extrinsics, encoder offsets).
Build health monitoring, FDIR , and graceful-degradation behaviors for harsh terrain and intermittent sensors.
Establish KPIs ( ATE/RTE , NEES/NIS , availability) and automated regression tests.
Tooling, Simulation & Field Operations
Create log-replay pipelines, datasets, and dashboards for rapid iteration and performance tracking.
Validate in sim (Gazebo, Isaac etc) and in the field (stairs, rubble, ramps, slippery floors).
Optimize and deploy on embedded targets (Jetson/x86), profiling latency, memory, and numerical stability.