進化戦略に基づく粒子フィルタによる状態推定(<小特集>制御工学への知能科学からの接近)  [in Japanese] State Estimation by Evolution Strategies Based Particle Filter(<Special Feature>Intelligent Systems Science Approach to Control Systems Science)  [in Japanese]

State estimation of dynamical systems is one of the most important problems in control systems science. This problem can be solved by a Bayesian approach, that is, inference on the unknown state can be performed according to the posterior probability distribution (pdf), which is obtained by combining a prior pdf for the unknown state with a likelihood function relating them to the observations. Kalman filter provides the optimal solution for linear dynamical systems under linear observation system with Gaussian noise. However, state space models in many realistic problems include nonlinear and non-Gaussian elements that preclude a closed form of expression for the optimal state estimate and then approximations are required. Here, several approximate filters are presented such as extended Kalman filter, Gaussian sum filter, point-mass filter and particle filters including a novel particle filter called Evolution Strategies Based Particle Filter (ESP), which has been developed by recognizing the similarities and differences of the operations in the particle filter and Evolution Strategies.