Semantic Distributed Multi-Robot Classification, Localization, and Mapping With a Viewpoint Dependent Classifier Model Supplementary Material

This formulation allows us to isolate the new information sent by other robots at time k, from information already used for inference at previous times. Next, we have to address that not all known objects are present in the sent local beliefs. Because the priors are assumed independent between poses and classes, P(X k \xk|C k ) = P(X k \xk). From P(X k \xk|C k ,H k−1) we can split X k \xk into poses of objects that are involved in H k−1 and ones that do not as: P(X k \xk|C k ,H k−1) = P(X o,R new,k|C R new,k,H k−1)P(X k−1|C k ,H k−1). (5)