An approach to integrating HLA federations and genetic algorithms to support automatic design evaluation for multi-agent systems

We propose a novel design environment for developing multi-agent systems (MASs) for applications in mobile robotics. Because emergent behavior phenomena make it next to impossible to directly synthesize viable MAS designs from specifications, extensive simulation studies are needed to evaluate these designs. Furthermore, due to the fact that the design space for MASs systems is combinatorially large, the evaluation of candidate designs must be done in a hierarchical, multi-resolution, parallel and distributed manner.

Our proposed design environment is based on US Department of Defense's high-level architecture (HLA), an established software infrastructure for heterogeneous, distributed simulations. The proposed environment automatically generates and manages HLA federations (i.e., collections of distributed and/or parallel simulation and service federates) that communicate over runtime infrastructure (RTI) software buses. Each federation simulates a different candidate design for the MAS under development. Federations execute independently and in parallel. Our proposed design environment's refinement component uses a genetic algorithm (GA) to select the best candidate designs from the current generation and generates a set of refined, next-generation, candidate designs. A federation is created and managed for each of the next-generation designs and the automatic design process is repeated.