Learning Bayesian Network Structure from Environment and Sensor Planning for Mobile Robot Localization
Abstract:
| In this paper we propose a novel method to solve a kidnapped robot localization problem. A mobile robot plans its sensing action for localization using learned Bayesian network's inference. Concretely, we represent the contextual relation between the local sensing results, actions and the global localization beliefs using the Bayesian network. The Bayesian network structure is learned from complete environment information data using K2 algorithm combined with GA. The mobile robot actively plans its sensing action to obtain sensing information event by taking into account the trade-off between global localization belief and the sensing cost. We have validated the learning and planning algorithm by simulation experiment in an office environment. |