Fuzzy Interpretation of Sequences of Multi-sensor Data for an Autonomous Mobile Robot

EXTENDED ABSTRACT Data interpretation is one of the most relevant issues to support autonomy in robotic, mobile agents. The integration of data from diierent sensors and the study of their temporal evolution may increase the reliability of the models used to control the agent. In this paper, we present an approach to interpret in real time data sequences provided by diierent sensors. Our system interprets data clusters in terms of fuzzy sets and uses this interpretation as a basis to recognize more complex situations both from the co-occurrence of classiication patterns provided by diierent sensors, and from sequences of classiication patterns by the same sensor. Thus, we can build models of the environment based on terms similar to those adopted by the designer of the control system to reason about the task. This reduces the design cost and increases the quality of the control system. We identify classes of homogeneous sensors, that is sensors that share some data characterisic, and we deene for each of them: how to characterize the terms we are interested in, how to map sensor data to these terms, and how to operate with the identiied terms. In the application we present in this paper, we have considered three sensor classes, concerning data from the odometer (the device measuring the agent's movement), from sonar sensors, and from an omnidirectional image sensor (COPIS 5). In this paper we focus on sonar data. We have implemented a system that can produce 8 sonar readings every 150 ms 3. This data rate makes it possible to identify relevant aspects of the temporal evolution of the readings. We cluster short sequences of readings from each sensor and we interpret them in terms of fuzzy sets. We have two main reasons to adopt this approach: rst, sonar data are aaected by errors, and their fuzzy interpretation is more robust than a crisp one; second, the identiication of more complex terms, based on the simpler ones, is more robust 1