Kálmán filter

Kálmán filter
Name	Kálmán filter
Inventor	Rudolf E. Kálmán
Introduced	1960
Field	Control theory, signal processing

Kálmán filter

The Kálmán filter is an algorithm for sequential state estimation in linear dynamical systems with Gaussian noise, widely used in NASA, Boeing, Siemens, Bell Labs and MIT research and applications. It combines prior state predictions with noisy measurements to produce minimum-variance estimates, and underpins technologies developed at Jet Propulsion Laboratory, Lockheed Martin, Honeywell, General Motors and Stanford University. The algorithm is foundational in modern IEEE control and signal processing curricula and appears in standards used by European Space Agency and NATO projects.

Introduction

The filter addresses discrete-time estimation for systems modeled by linear stochastic difference equations, integrating inputs from devices such as sensors devised by Raytheon, Texas Instruments, Philips, Panasonic and Bosch. It is central to navigation systems designed by Garmin, Magellan, Thales Group and Northrop Grumman, and forms the backbone of state estimation routines in flight control at Airbus and Lockheed Martin. The technique is also applied in econometric forecasting practiced at Federal Reserve Board, International Monetary Fund, World Bank and research groups at Princeton University and University of Chicago.

Mathematical Formulation

Consider a linear discrete-time system with state vector x_k and measurement y_k; dynamics are given by x_{k+1} = A_k x_k + B_k u_k + w_k and y_k = C_k x_k + v_k, where A_k, B_k, C_k are matrices estimated in labs at Bell Labs and IBM Research, u_k is a control input used by teams at DARPA and US Air Force, and w_k, v_k are zero-mean Gaussian noises with covariances Q_k and R_k respectively characterized in experiments at Lawrence Livermore National Laboratory, Argonne National Laboratory and Los Alamos National Laboratory. The optimality condition follows from minimizing the expected squared estimation error E[(x_k − x̂_k)(x_k − x̂_k)ᵀ] under assumptions commonly employed in analyses at Caltech, ETH Zurich and Imperial College London. The steady-state solution for time-invariant systems uses algebraic Riccati equations familiar to researchers at Princeton University and developers at Toyota Research Institute and IBM.

Algorithm Implementation

The recursive implementation alternates a prediction step and an update step, operations used in software stacks from Microsoft Research, Google robotics teams, and Apple sensor fusion libraries. Prediction propagates state and covariance using A_k and Q_k; update computes the Kalman gain from predicted covariance and measurement noise R_k, incorporating incoming y_k from hardware by Sony, Nokia, Samsung and Qualcomm. Numerical stability techniques such as Joseph form, square-root filtering, and UD factorization are adopted in implementations by NVIDIA, Intel, AMD, ARM Holdings and control toolboxes in MATLAB and GNU Octave. Real-time implementations in embedded systems appear in products by STMicroelectronics, Microchip Technology, Texas Instruments and Xilinx.

Variants and Extensions

Extensions include the Extended filter used for nonlinear models studied at University of California, Berkeley, the Unscented variant developed in work affiliated with Georgia Institute of Technology and the Ensemble filter popularized in climatology at National Oceanic and Atmospheric Administration and Met Office. Robust and adaptive filters incorporate concepts from teams at University of Cambridge, Imperial College London and University of Oxford; constrained, information, and square-root forms appear in publications from IEEE Control Systems Society, SIAM, and ACM. Particle filtering methods, advanced at Columbia University and University of Toronto, generalize the framework for non-Gaussian, nonlinear settings used in research at ETH Zurich and University College London.

Applications

The filter is deployed in inertial navigation for spacecraft at European Space Agency and NASA missions, in autonomous vehicles developed by Waymo, Cruise, Tesla, and in robotics by Boston Dynamics and iRobot. It aids target tracking in systems from Raytheon and Northrop Grumman, satellite orbit determination at SpaceX and ULA, and sensor fusion in smartphones by Apple and Samsung. Econometric uses occur in forecasting at Federal Reserve Board and OECD, while biomedical signal processing applications are pursued at Johns Hopkins University, Mayo Clinic, and Massachusetts General Hospital.

Performance and Limitations

Under linearity and Gaussian noise assumptions, the filter yields minimum-variance unbiased estimates, a property proven in contexts developed by researchers at Princeton University, Harvard University, Yale University and Stanford University. Performance degrades with model mismatch, non-Gaussian noise, and severe nonlinearity encountered in projects at DARPA and Defense Advanced Research Projects Agency-sponsored initiatives; in such cases, particle or ensemble methods from Los Alamos National Laboratory and NASA Jet Propulsion Laboratory are preferred. Numerical instability and computational cost issues arise in high-dimensional settings addressed by teams at Google DeepMind, OpenAI, Facebook AI Research and supercomputing centers like Oak Ridge National Laboratory.

Historical Development and Naming

The theory was formalized in a landmark 1960 paper by Rudolf E. Kálmán while collaborating with researchers at Massachusetts Institute of Technology and Princeton University, with precursor ideas appearing in work by Norbert Wiener and Andrey Kolmogorov from institutions including Yale University and University of Moscow. Early adoption in aerospace engineering was driven by engineers at NASA and JPL and in guidance systems at Lockheed Martin and Boeing. Subsequent theoretical and practical developments were advanced by scholars affiliated with University of California, Berkeley, Cornell University, University of Michigan, Caltech and Stanford University.

Category:Estimation theory