Abstract

In this work, a novel methodology for robots executing informed object search is proposed. It uses basic spatial relations, which are represented by simple-shaped probability distributions describing the spatial relations between objects in space. Complex spatial relations can be defined as weighted sums of basic spatial relations using co-occurrence matrices as weights. Spatial relation masks are an alternative representation defined by sampling spatial relation distributions over a grid. A Bayesian framework for informed object search using convolutions between observation likelihoods and spatial relation masks is also provided. A set of spatial relation masks for the objects monitor, keyboard, system unit and router were estimated by using images from Label-Me and Flickr. A total of 4,320 experiments comparing six object search algorithms were realized by using the simulator Player/Stage. Results show that the use of the proposed methodology has a detection rate of 73.9% that is more than the double of the detection rate of previous informed object search methods.