A K-best Filtering k-1 k-1) 12 component I component 2 compone transition expansion estimation A* search through space of possible successors [h&W 2002] Evaluated in the order of f(n,)=g(n1)+h(n,) Admissible heuristics Probability(cost)of trajectory so far upperbound on probability from n. onward Hybrid Mode Estimation and Gaussian Filtering with Hybrid HMM Models 16412/6.834 Lecture,15 March2004 Rao-Blackwellised Particle Filtering Principle: Decrease the computational complexity by reducing dimensionality of sampled space Sample variables r Closed-form solution for variables s Hybrid Mode Estimation and Gaussian Filtering with Hybrid HMM Models 6412/6.834 Lecture,15 March2004Hybrid Mode Estimation and Gaussian Filtering with Hybrid HMM Models 16.412 / 6.834 Lecture, 15 March 2004 31 K-best Filtering z A* search through space of possible successors [H&W 2002] z Evaluated in the order of ( ) ( ) ( ) nQ g nQ h nQ f  Probability (cost) of trajectory so far Admissible heuristics: upperbound on probability from nv onward Hybrid Mode Estimation and Gaussian Filtering with Hybrid HMM Models 16.412 / 6.834 Lecture, 15 March 2004 32 Rao-Blackwellised Particle Filtering Principle: Decrease the computational complexity by reducing dimensionality of sampled space ƒ Sample variables r ƒ Closed-form solution for variables s r s r ( | ) ( ) 0: i t p s r s component 1component 2 PT2 n v PT1 PT1 P o h(k) h(k-1) (k-1) transition expansion estimation component l x x(k)