A multi-agent societal architecture for real-time applications

The design of complex real-time systems is a difficult task. There are a number of fundamental aspects that are important when considering such systems. Firstly, they must be resilient to errors or plan deviations. Secondly, such systems must be capable of being designed and constructed in an evolutionary manner. Thirdly, the final delivered implementation must meet the timing constraints imposed by the environment. An integrated approach is necessary to tackle these problems. This integration must encompass the total hardware/software development environment as well as the integration of the sub-units of a complex system solution. The application that has driven this research has been based on the building of an assembly cell using multiple robots sharing a common workspace with multiple sensors tracking the dynamic state of the working environment. The multiple robots provide flexibility and reduce the level of specialised fixturing by acting as programmable fixtures in addition to carrying out assembly operations. A brief overview is presented of a societal architecture which will support the structures that are likely to be met in complex real-time systems. The authors have developed a design methodology for this type of architecture and have applied it to the control of a multi-robot-multi-sensor assembly workcell where more than 50 agents and over 200 frames of data carry out the control task. They found that the architecture is extendible and robust and that the methodology works well in the evolutionary software environment