Decentralised Data Fusion Using Agents
Charles R. Robinson · 2008
This project investigates aspects of the dynamic control problem where, for example, timeliness, robustness and fault tolerance are critical. An objective is to produce a non-hierarchical structure with the potential for emergent intelligent behaviour, such as would benefit an autonomous vehicle system. Following a general introduction, there is an extensive and detailed discussion of the theories and ideas which support the concept of decentralised data fusion, and the use of agents, while also providing the rationale for subsequent work. This leads to the development of two models. One of these acts as a test-bed to provide data for subsequent analyses. The other is a model which implements data fusion using agents. The two models are linked through an interface called MACSim. MACSim is developed to enable communication between Simulink and JADE, thus bringing together two powerful software tools for modelling the hardware of real-time systems and multi-agent systems. This will facilitate future research and development on decentralised data fusion systems using multiple agents. In the present project a Boeing 747 sensor system is modelled using Simulink. Boeing 747 sensor data is then used to test the operation of an agent-mediated decentralised data fusion system. A variety of experiments are described with detailed discussion of the results obtained. The findings demonstrate the successful implementation of Simulink and JADE operating together. In addition, the results confirm that the performance of a decentralised fusion process, using information filters, can be comparable with a centralised fusion process using Kalman filters. Throughout the testing process there is comparison of three different forms of information filter. The report concludes with a detailed discussion of implications for future research and for further development of the models employed.