Multi-Modal Particle Filtering for Hybrid Systems with Autonomous Mode Transitions

Abstract

Model-based diagnosis of embedded systems relies on the ability to estimate their hybrid state from noisy observations. This task is especially challenging for systems with many state variables and autonomous transitions. We propose a fair sampling algorithm that combines Rao-Blackwellised particle filters with a multi-modal Gaussian representation. In order to handle autonomous transitions, we let the continuous state estimates contribute to the proposal distribution in the particle filter. The algorithm outperforms purely simulational particle filters and provides unification of particle filters with hybrid hidden Markov model (HMM) observers.