What Are Observers?

Observers, in control systems engineering, are algorithms that estimate the internal state of a dynamical system using a model of the system's behavior and a stream of measured outputs. Because physical systems often have more state variables than available sensors, many of the quantities needed to compute a control action are not directly measurable. Observers reconstruct those unmeasured states in real time, making full-state feedback control feasible where direct measurement would be costly, impractical, or technically impossible. The theoretical basis for observers was established by David Luenberger in 1964 and Rudolf Kalman in the early 1960s, and the concept has since been generalized to nonlinear and stochastic systems.

Luenberger Observers

The Luenberger observer, also called a deterministic state observer, is designed for linear time-invariant systems. It maintains an internal simulation of the plant and adds a correction term proportional to the difference between the measured output and the output predicted by the simulation. This output injection term drives the observer state toward the true system state, with a rate determined by the observer gain matrix. The observer poles, which govern how quickly the estimation error decays, are placed by choosing the gain matrix to achieve a desired closed-loop eigenvalue spectrum; a common design rule is to place observer poles three to five times faster than the slowest closed-loop plant poles. The classic paper An introduction to observers by David Luenberger himself, published in IEEE Transactions on Automatic Control, set out the framework that remains standard in linear control curricula and practice. Reduced-order Luenberger observers estimate only the unmeasured states, exploiting the fact that measured outputs are already known without reconstruction.

Kalman Filter Observers

The Kalman filter is the optimal state estimator for linear systems driven by Gaussian process noise and corrupted by Gaussian measurement noise. It generalizes the Luenberger observer by treating state estimation as a minimum-variance problem: at each time step, the filter computes a gain matrix that balances the uncertainty in the dynamic model against the uncertainty in the measurement. The result is a recursive algorithm that produces a state estimate along with an error covariance matrix quantifying estimation confidence. The Extended Kalman Filter (EKF) applies this framework to nonlinear systems by linearizing the dynamics around the current estimate, and the Unscented Kalman Filter (UKF) improves on the EKF by propagating a carefully chosen set of sigma points through the full nonlinear equations rather than a linear approximation. A Luenberger state observer for simultaneous estimation of speed and stator resistance in sensorless induction motor drives illustrates how observer designs adapted from the Kalman framework are implemented in demanding industrial applications.

Nonlinear and Sliding Mode Observers

Nonlinear observers extend the estimation problem to systems whose dynamics cannot be accurately represented as a linear model. The high-gain observer uses large gains in the output injection term to make the estimation error dynamics fast relative to the system's nonlinear terms, exploiting a time-scale separation. Sliding mode observers use a discontinuous switching correction term that drives the output estimation error to zero in finite time; the resulting sliding surface then constrains the observer state to track the true state exactly in the absence of noise. These observers are valued in applications where disturbances and model uncertainties are large because the sliding condition ensures exact output matching despite bounded perturbations. In machine vision applications, observers estimate the pose and motion of tracked objects, functioning as the state estimation layer in visual servo control loops where camera measurements replace traditional sensors. An analysis at Rutgers University's Introduction to Linear and Nonlinear Observers surveys the theoretical connections between these designs and the controllability-observability duality that underlies all of them.

Applications

Observers have applications in a wide range of fields, including:

• Electric motor drives and power electronics, for sensorless speed and flux estimation
• Aerospace and navigation, supporting inertial navigation and fault-tolerant flight control
• Machine vision and visual servoing, where camera measurements feed observer-based pose estimation
• Process control in chemical plants, estimating unmeasured concentrations from temperature and flow sensors
• Fault detection and isolation, by comparing observer-predicted outputs to measured values and flagging deviations