Maximum Likelihood Fusion of Stochastic Maps

Abstract

The fusion of independently obtained stochastic maps by collaborating mobile agents is considered. The proposed approach includes two parts: generalized likehood ratio matching and maximum likelihood alignment. In particular, an affine invariant hypergraph model is constructed for each stochastic map and a bipartite matching via a linear program is used to establish landmark correspondence between stochastic maps. A maximum likelihood alignment procedure is proposed to estimate rotation, translation and scale parameters in order to construct a global map of the environment. A main feature of the proposed approach is its scalability with respect to the number of landmarks: the matching step has polynomial complexity and the maximum likelihood alignment solution is obtained in closed form. Experimental validation of the proposed fusion approach is performed using the Victoria Park experimental benchmark.