Reverse KLD-Sampling for Measuring Uncertainty in Rao-Blackwellized Particle Filters SLAM
Luca Carlone, Miguel Efrain Kaouk Ng, Jingjing Du, Basilio Bona, Marina Indri · PORTO Publications Open Repository TOrino (Politecnico di Torino) · 2009
( ) / 0 0 0 obs obs I EM N if N EMMI otherwise > > = The information I(EM) can be computed from the entropy of the expected map, whereas the EMMI is obtained including the number of observed cells Nobs that are present in the map: We assume that the quality of map building process depends on the approximation of the position belief. If particles are drawn from a discrete distribution, with support over kt bins, at each time step we compute the probability 1-t of having a divergence less than between the true distribution p and our particle-based representation of pose belief xt : p