This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| EKF | |
|---|---|
| Name | EKF |
| Caption | Extended Kalman Filter schematic |
| Invented by | Rudolf E. Kálmán |
| Introduced | 1960s |
| Field | Control theory; Signal processing |
| Applications | Apollo program; Global Positioning System; Inertial navigation system; Autonomous vehicle; Robotics |
EKF
The Extended Kalman Filter (EKF) is a recursive state estimator for nonlinear dynamical systems that linearizes dynamics and measurement models around current estimates to extend the Kalman filter framework to a broad set of problems in Aerospace and Robotics. It combines prior information, sensor measurements, and process models to produce minimum-variance estimates under Gaussian assumptions, and has been used in projects such as the Apollo program and implementations in Global Positioning System receivers and Inertial navigation system suites. EKF bridges classical estimation theory from Rudolf E. Kálmán with practical implementations in systems designed by organizations like NASA and DARPA.
EKF originated as an adaptation of the Kalman filter to accommodate nonlinear state-transition and observation functions encountered in systems developed by NASA and the aerospace community. EKF is widely taught in courses at institutions such as Massachusetts Institute of Technology, Stanford University, and University of California, Berkeley and features in textbooks by authors like Dan Simon and G. F. Franklin. It underpins navigation stacks in platforms by companies such as Google (Waymo) and Tesla, Inc. and appears in research from labs at MIT CSAIL and Stanford Artificial Intelligence Laboratory.
Consider a discrete-time nonlinear system with state x_k and observation z_k modeled by functions f and h: - State prediction: x_k = f(x_{k-1}, u_{k-1}) + w_{k-1} - Observation: z_k = h(x_k) + v_k
Here process noise w and measurement noise v are modeled as zero-mean Gaussian with covariances Q and R, akin to formulations used in studies at Princeton University and Caltech. EKF linearizes f and h about the current estimate using Jacobians F_k = ∂f/∂x and H_k = ∂h/∂x, paralleling linearization techniques applied in work by Arthur E. Bryson and Bruce A. Francis. The prediction step propagates mean and covariance; the update step applies a Kalman gain K_k derived from linearized observation Jacobian H_k, consistent with derivations in texts from IEEE journals and conferences such as ICASSP and CDC.
The EKF algorithm alternates predict and update phases. In predict: propagate the state via f and the covariance via F_k and Q. In update: compute residual y_k = z_k − h(x_k^-), form S_k = H_k P_k^- H_k^T + R, compute K_k = P_k^- H_k^T S_k^{-1}, and update x_k and P_k. Implementations appear in software libraries from ROS communities, Eigen (C++ library), and toolkits in MATLAB and GNU Octave. Numerical stability techniques include Joseph form covariance updates used in projects by Lockheed Martin and Northrop Grumman, and square-root implementations inspired by algorithms in Stanford Linear Accelerator Center literature.
Numerous variants extend EKF to address nonlinearities, non-Gaussianity, and computational constraints. The Iterated EKF (IEKF) refines linearization via repeated measurement linearizations, an approach applied in research at Imperial College London and ETH Zurich. The Unscented Kalman Filter (UKF) from Simon Julier and Jeffrey Uhlmann avoids explicit Jacobians using sigma points; it is often compared alongside ensemble Kalman filters (EnKF) used in atmospheric science at NOAA and European Centre for Medium-Range Weather Forecasts. Augmented-state EKF handles bias estimation in inertial sensors as used by Honeywell and Raytheon. Multi-sensor fusion architectures combine EKF with particle filters as in systems developed at Carnegie Mellon University and MIT Lincoln Laboratory.
EKF has been deployed across aerospace, marine, automotive, and consumer electronics. In spacecraft navigation it aided Apollo program guidance and remains in variants for missions by ESA and JAXA. In aviation, EKF is integrated into flight management systems by manufacturers such as Boeing and Airbus. In robotics and autonomous vehicles, EKF fuses LiDAR, camera, and IMU data in projects at Waymo, Cruise (company), and research from Oxford Robotics Institute and ETH Zurich. EKF is used in smartphone sensor fusion in products by Apple Inc. and Samsung Electronics, and in geodesy and surveying with receivers from Trimble and Topcon.
Under local linearity and Gaussian noise assumptions, EKF provides near-optimal minimum-variance estimates and has low computational cost relative to sampling-based methods used at Google Research and in particle filter literature from University of Washington. Limitations include sensitivity to initial conditions, divergence under strong nonlinearities, and incorrect covariance estimates, problems documented in case studies at NASA Jet Propulsion Laboratory and by researchers at UC San Diego. Compared to UKF and EnKF, EKF requires analytical Jacobians which can be burdensome in complex models developed at Siemens and ABB. Particle filters handle multimodal distributions in scenarios studied at Rutgers University but at higher computational expense compared to EKF.
Real-world EKF deployment requires careful modeling of Q and R, state augmentation for sensor biases, and observability analysis as practiced in projects at MITRE Corporation and Ames Research Center. Tuning methods include covariance inflation, adaptive noise estimation used in NOAA data assimilation, and residual monitoring techniques employed by FAA avionics certification teams. Debugging tools include consistency tests like normalized innovation squared (NIS) referenced in IEEE Transactions on Aerospace and Electronic Systems, and Monte Carlo validation workflows used in simulation platforms by Siemens PLM and Ansys.
Category:Estimation theory