Optimal Intermittent Particle Filter

The problem of the optimal allocation (in expected mean square error sense) a measurement budget for particle filtering is addressed. We propose three different intermittent filters, whose optimality criteria depend on information available at time decision making. For first, stochastic program filter, times are given by policy that determines whether should be taken based measurements already acquired. second, called offline all once solving combinatorial optimization before any acquisition. third one, which we call online each new received, next recomputed to take then into account. prove in terms error, filter outperforms itself filter. However, these filters generally intractable. this reason, estimate approximated Moreover, using Monte-Carlo approach, and algorithms compared approximately solve programs (a random trial algorithm, greedy forward backward algorithms, simulated annealing genetic algorithm). Finally, performance proposed methods illustrated two examples: tumor motion model common benchmark filtering.