The Unscented Kalman Filter (UKF) is a recursive Bayesian estimation algorithm that extends the classical Kalman Filter to nonlinear systems by propagating a carefully chosen set of deterministic sample points, called sigma points, through the nonlinear state transition and observation functions to approximate the posterior mean and covariance without requiring explicit Jacobian computation. In the context of traffic state estimation and agent-based pedestrian modeling, the UKF is valued for its ability to fuse noisy sensor observations with dynamic system models in real time, providing more accurate and stable state estimates than the linearization-based Extended Kalman Filter, particularly when system dynamics are highly nonlinear. Key variants include the square-root UKF, which improves numerical stability, and ensemble-based adaptations that connect the UKF to broader data assimilation frameworks such as the Ensemble Kalman Filter, enabling scalable application to high-dimensional agent-based and traffic flow models.
Dictionary term