Mission Management for Multiple Autonomous Vehicles

摘要

This article discusses research into the engineering of Mission Management Knowledge Based processes for the command of Multiple Intelligent Autonomous Vehicles(MIAVs). In particular it discusses architectural and algorithmic considerations in the light of the demanding requirements for robustness and increased system longevity. The architectural issues covered reflect recent developments in Object Technology which has demonstrated the benefits of a componentised view of systems, where observation of interface standards can provide for a 'plug and play' approach to system development and evolution. The algorithmic considerations concentrate on the significant progress in the machine learning field specifically looking at combining popular Knowledge Based Systems(KBSs) approaches with those developed in the area of Artificial Neural Networks (ANNs). Adaptive systems promise resistance to change through the modification of internal models as a result of direct experience of the problem domain, and can, under certain conditions, behave robustly in unseen situations. The engineering of Mission Management Knowledge Based processes for the command of Multiple Intelligent Autonomous Vehicles(MIAVs) concerns the coordination of elements of a distributed system so as to generate coherent behaviour. As such the techniques apply to the management and control of envisaged civil information and automation systems in public utilities, transportation and manufacturing as well as military command and control. The ever increasing demand for cost effectiveness, project efficiency and increasing productivity resulting from open competition is resulting in greater demand for coherent, systems solutions for bespoke large scale projects, such as major building construction, air traffic control and road management systems. These integrated systems are characterised by high capital value and extended life time, which together raise a requirement for evolution in system capability and the acceptance of change as an inherent characteristic of system infrastructure. The acceptance of change implies the need for a rigorous approach to the analysis and design of these systems which emphasises the achievement of modularity. In addition, since these systems often are required to operate in dynamic large, complex, uncertain, unstructured, non-benign environments without human intervention, there is a requirement for an intelligent adaptive ability which can react to environmental dynamics. Adaptive systems are more resistant to system and environmental changes potentially resulting in significant cost saving though increased operational life. The engineering of Mission Management systems for Multiple Intelligent Autonomous Vehicles(MIAVs) in particular and the management problem domain in general places heavy reliance on human decision making and supervision. Computerised management systems have been difficult to introduce primarily as a result of inadequacies in the technology. This is, in part, due to difficulties with describing models of the domain with sufficient precision. Experts in management have a good 'feel' for problems in the domain but, despite being effective managers, find it difficult to express their knowledge in anything but an approximate, vague, rule of thumb way. This conflicts with computer systems requirements which need a precise and complete description of domain relationships. Robotic computerised management solutions traditionally involve the Knowledge Based Planning(KBP) of activities over time and their monitoring during execution. This research extends KBP to approximate rule based systems to support initialisation from expert knowledge, while supporting adaption to fine tune approximate rule sets to better describe domain relationships. This approach takes advantage of expressible human expertise while compensating for inaccuracy, ignorance and incompleteness by supporting adaption, giving systems increased resistance to chang