Cooperative Topological Map Building Using Distributed Consensus

Abstract

“A good map is both a useful tool and a magic carpet to far away places.” We have studied how to modify the consensus iteration to handle different perception issues. In this chapter we present an application of such algorithms in the problem of cooperative mapping with cameras. The approach builds topological maps from the sequences of images acquired by each robot, grouping the features in planar regions and fusing them using consensus. The use of planar regions to represent the map has many advantages both in the mapping task and in the achievement of the consensus. First of all, using inter-image homographies, the individual maps are easy to create and the data association between different maps is simple. The computation of a global reference frame to represent the features, which is in general quite complicated, but necessary to reach a consensus, is reduced to a simple max-consensus method multiplying different homographies. Finally, homographies between images can be computed without knowing the internal parameters of the cameras, which makes the approach robust to calibration issues. The result is a simple but very effective distributed algorithm that creates a global map using the information of all the robots. Experiments with real images in complex scenarios show the good performance of the studied distributed solution.