A successful design optimization depends on the underlying representation. Complex system engineering addresses the challenges of the optimal representation through key features like self-organization, modularity, or locality. The representation covers the parameter setup (location and quantity) and the mapping between parameter space (genotype) and design space (phenotype), and should allow for both adaptation and specialization of a design. To quantify the potential of a representation, suitable quality criteria are needed. Evolvability is such a criterion. In biology, evolvability characterizes the potential success of a population. We analyze, interpret, and extend definitions of evolvability in order to derive an evolvability criterion suitable for complex system engineering. The first aspect that evolvability has to cover is regularity to prevent unfavorable designs. The second aspect is variability to react to changing conditions. Thirdly, it is important that the representation promotes those regions that are favorable to the design process.
System Optimization describes the process of determining the optimal structure, parameterization and adaptation of systems situated in complex environments. The interdisciplinarity of engineering design, the high connectivity of large-scale economic models or the multi-scale complexity of intelligent systems are examples for the challenges of System Optimization. Like in Systems- and Requirements-Engineering the whole life-cycle has to be optimized. The optimal spatio-temporal decomposition into subsystems, their patterns of interaction and the propogation of uncertainties are central questions.
At HRI we approach System Optimization with stochastic optimization methods inspired by biological evolution. Biological evolution is situated design or optimization in-vivo. This means that the design and the operation process are occuring concurrently in the same environment. In our research we focus on the following aspects of system optimization:
• Many objective and many disciplinary optimization
• Robustness as an inherent property of the structure of systems
• Evolvability that enables systems to continuously adapt to dynamic environments
• MDS (modularize, duplicate, specialize) principle for incremental system design
For more information
A. Richter, M. Botsch and S. Menzel, “Evolvability of representations in complex system engineering: A survey”,
Proc. IEEE Congr. Evol. Comput. (CEC), 2015, pp. 1327-1335.