Ricerc@Sapienza

This paper presents a benchmark for object recognition inspired by RoboCup@Home competition and thus focusing on home robots. The benchmark includes a large-scale training set of 196K images labelled with classes derived from RoboCup@Home rulebooks, two medium-scale test sets (one taken with a Pepper robot) with different objects and different backgrounds with respect to the training set, a robot behavior for image acquisition, and several analysis of the results that are useful both for RoboCup@Home Technical Committee to define competition tests and for RoboCup@Home teams to implement effective object recognition components.