A multi-robot system : from design to intelligent cooperation

Abstract

The proposed work aims at giving a contribution in the domain of multi robot systems. In the last years these architectures, thanks to their flexibility and robustness with respect to monolithic structures, have been deeply investigated in various applications. Looking at commercial platforms, one of the main limitations relies on the hard- ware configuration. In most of cases, the low-level control software is not open source, compromising a detailed customization of the robot. This drawback can be partially overcome by the usage of plug-in boards which ease the use of a particular sensor. How- ever, whenever an application requires heterogeneous transducers, the management of several boards increases the computational burden of the CPU, the energy consumption and costs. The first contribution of this thesis is the description of the prototyping of a multi robot platform. The robots, based on a dual layer architecture, have been designed to be low-cost and flexible in order to cope with several scenarios. The reduced di- mensions allow the units to execute experiments in small environments, like the ones typically available in a research laboratory, even interacting with commercial platforms like sensor networks. Afterward, several experimental research activities are covered. In the field of motion control, two techniques for the encircling problem, i.e. the task in which a set of robots rotates around a target, are presented. This toy problem is very appealing to the scientific community because it deals with a series of technological issues which involve communications, motion estimation and control design. Also the data fusion, which is one of the most treated problem in robotics, has been taken into account. One of the traditional approaches resides on the Bayesian framework. This approach allows a straightforward combination of data gathered by several sensors. In this dissertation, such technique has been used in environmentalmonitoring. In particular extensive experiments were performed using the aforemen- tioned platforms in the domain of gas source localization: the task consists in the search of a source releasing gas in an environment. Besides the Bayesian framework, other techniques can be used. In particular a focus on the Transferable Belief Model (TBM) is provided. Unlike the probabilistic formulation, the TBM is able to effectively repre- sent the concept of ignorance and contradiction, concepts which, for example, belong to the human kind of thinking. In this work a technique for the data fusion in a multi agent context is proposed. The theoretical results have been validated by extensive experiments concerning the cooperative topological map building. Finally a lookup view about the integration between mobile and static sensor net- works (SN) is provided. One of the main issues for a static network is to know about the interdistances between nodes. In this work a localization algorithm, based on a modified Extended Kalman Filter, is showed. As for the previous topics, experimental results support the theoretical approach. In the last chapter conclusions and possible improvements are discussed.