Efficiently exploring for human robot interaction: partially observable Poisson processes

Abstract Consider a mobile robot exploring an office building with the aim of observing as much human activity possible over several days. It must learn where and when people are to be found, count observed activities, revisit popular places at right time. In this paper we present series Bayesian estimators for levels that improve on simple counting. We then show how these can used drive efficient exploration activities. The arise from modelling counts partially observable Poisson process (POPP). This presents novel extensions POPP following cases: (i) robot’s sensors correlated, (ii) sensor model, itself built data, is also unreliable, (iii) both combined. combines resulting simple, but effective solution exploration-exploitation trade-off faced by in real deployment. A 15 day deployments our approach boosts number activities 70% relative baseline produces more accurate estimates level each place