The first version of Agent-Space architecture (called Agent-Environment) has been introduced by Andrej Lucny in 1997 as a proposal for his doctoral thesis, aimed to express Brooks' subsumption architecture by language of multi-agent systems, i.e. to leave its hardware-oriented fashion and give it pure software nature. Another motivation was looking for an architecture for multi-agent system building where each agent could be recursively conceived as multi-agent system of the same fashion but of a lower level. This idea has been inspired by principle proposed by supervisor Jozef Kelemen: that hierarchy of agent can be turned to encapsulation within single level.
The first implementation was developed in C under QNX4 real-time OS in 1997. QNX4 was a wonderful platform which - due to its philosophy of microkernel and SRR message passing - enabled to run lot of programs exchanging data on one hand but endangered such solution by various possibilities for deadlocks and livelocks. Thus it was necessary to establish clear rules how build such multi-process system - in other word to specify an architecture. Rather than rules for establishing an order, Agent-Space architecture disabled any chance for deadlock or livelock in principle, requiring no order at all. It has been used for several industrial applications (mostly monitoring systems in MicroStep-MIS, including system for precise measurement of thermal norm for nuclear power reactor) and have brought reliability of these products to better qualitative level, mainly due to simplification of so-called soft-crash landing (automatic restart of crashed processes). Unfortunately, due to the fact that QNX4 has not been ported to modern hardware, nowadays, this implementation can be launched just under virtual machines.
Soon it was tried to use the architecture also for academic purposes which was promising more intensive requirement for manifold hierarchy of agents. The first trial have been implemented under QNX implementation in 2001 (modeling of digger wasp mating behavior).
In 2001 it was found that many ideas of the architecture have been invented long-time ago in domain of parallel programming, namely idea of LINDA space (1985). Therefore we renamed the architecture to Agent-Space.
Another stimulus for further development came from initiative robotics.sk in 2003. The architecture has been tested for building control system of robots. The first trial (robot Jori) has been performed in 2003, but for summer camp of robotics.sk in 2004 we implemented agent-space architecture in Java (robot Pingpong). We used this implementation also for more serious research within DENSO project for pedestrian tracking from vehicle.
In 2009 we created this web-page, supported by grant VEGA 1/0280/08 "Application of multi-agent modularity for building of control systems and computer simulations".
Acknowledgements belongs mainly to my supervisor Jozef Kelemen who put me on this research and to my chief at Comenius University Jan Sefranek who enabled me to be active in academic sphere. Many thanks also to my Jozef Omelka, director of MicroStep-MIS, for his excellent and brave decision to employ this architecture in commercial products. Significant contribution to this project has been made by invention of my students at Comenius University, mainly to Karol Ostrovsky.