Robotics and Autonomous Systems 55 (2007) 253–265 www.elsevier.com/locate/robot On the use of Bayesian Networks to develop behaviours for mobile robots ✩ E. Lazkano ∗ , B. Sierra, A. Astigarraga, J.M. Mart´ ınez-Otzeta Department of Computer Science and Artificial Intelligence, University of the Basque Country, P.O. Box 649, E-20080 San Sebasti´ an, Spain Received 13 September 2005; received in revised form 20 July 2006; accepted 1 August 2006 Available online 26 September 2006 Abstract Bayesian Networks are models which capture uncertainties in terms of probabilities that can be used to perform reasoning under uncertainty. This paper presents an attempt to use Bayesian Networks as a learning technique to manage task execution in mobile robotics. To learn the Bayesian Network structure from data, the K2 structural learning algorithm is used, combined with three different net evaluation metrics. The experiment led to a new hybrid multiclassifying system resulting from the combination of 1-NN with the Bayesian Network, that allows one to use the power of the Bayesian Network while avoiding the computational burden of the reasoning mechanism — the so-called evidence propagation process. As an application example we present an approach of the presented paradigm to implement a door-crossing behaviour in a mobile robot using only sonar readings, in an environment with smooth walls and doors. Both the performance of the learning mechanism and the experiments run in the real robot-environment system show that Bayesian Networks are valuable learning mechanisms, able to deal with the uncertainty and variability inherent to such systems. c  2006 Elsevier B.V. All rights reserved. Keywords: Bayesian networks; Hybrid multiclassifiers; Mobile robots; Door-crossing behaviour; Behaviour-based systems 1. Introduction Bayesian Networks (BNs) are models which capture uncertainties in terms of probabilities that can be used, if correctly constructed, to perform reasoning under uncertain conditions [40,21,49,44]. Bayesian Network paradigm learning capabilities are exploited in very different domains with inherent uncertainty [24] like the detection of anomalies in Internet-based services [3]. In medical diagnosis, they have been used to help in the prediction of malignant skin melanoma, to help in the treatment of anaemia, and also for visual disturbances evaluation [48]; BNs are also applied in the field of fault diagnosis. For instance, in [23] a troubleshooting system for fixing faults in printing systems is presented, and [29] Lerner et al. model a complex hybrid system as a dynamic Bayesian Network. ✩ This work was supported by the Etortek AMIGUNE project, by the Gipuzkoako Foru Aldundi Txit Gorena under OF200/2005 grant and by the MCYT under grant TSI2005–00390. ∗ Corresponding author. Fax: +34 943015590. E-mail addresses: ccplaore@si.ehu.es (E. Lazkano), ccpsiarb@si.ehu.es (B. Sierra), ccbaspaa@si.ehu.es (A. Astigarraga), ccbmaotj@si.ehu.es (J.M. Mart´ ınez-Otzeta). URL: http://www.sc.ehu.es/ccwrobot. The property of uncertainty that sets inference in probabilistic networks apart from other automatic reasoning paradigms is its ability to make inter-causal reasoning: getting evidence that supports solely a single hypothesis (or a subset of hypotheses) automatically leads to decreasing belief in the unsupported, competing hypotheses. This property is often referred to as the explaining away effect. The ability of probabilistic networks to automatically perform inter-causal inferences, i.e. the probabilistic inference process, is a key contribution to their reasoning power [41]. Although Bayesian reasoning is widely used, especially for robot localisation [6,53], and other learning techniques such as Neural Networks [34] or Genetic Algorithms [43] are widely used, we found no references in the literature about the application of Bayesian Networks as such in the area of mobile robotics. We found no reason for this absence except for the difficulties of reasoning in real time due to the computational burden of the probabilistic inference process. Therefore, the main motivation of the experiment presented in this paper is to emphasise that the learning capabilities of Bayesian Networks can be used as a learning mechanism in mobile robotics. The BN paradigm has been tested for a sonar-based door- crossing behaviour that has been implemented using several 0921-8890/$ - see front matter c  2006 Elsevier B.V. All rights reserved. doi:10.1016/j.robot.2006.08.003