Online Joint State Inference and Learning of Partially Unknown State-Space Models

A computationally efficient method for online joint state inference and dynamical model learning is presented. The combines an a priori known, physically derived, state-space with radial basis function expansion representing unknown system dynamics inherits properties from both physical data-driven modeling. uses extended Kalman filter approach to jointly estimate the of learn dynamics, via parameters expansion. key contribution computational complexity reduction compared similar globally supported functions. By using compactly functions approximate gain, considerably reduced essentially determined by support approximation works well when exhibit limited correlation between points separated in domain. exemplified two intelligent vehicle applications where it shown to: (i) have competitive estimation performance method, (ii) be real-time applicable problems large-scale state-space.