Design.R - AI-Assisted CPS Design: Symbiotic Design for Cyber-Physical Systems

The goals of this Defense Advanced Research Projects Agency Symbiotic Design for Cyber-Physical Systems program were synthesizing, exploring, and optimizing multi-domain executable models with AI-based approaches. To mitigate the lack of large and coherent engineering models - using the same design tools, language, and format - several tools and technologies were developed for generative synthesis and automatic evaluation of models and conceptual design considerations. Most importantly, we developed a constraint programming framework, which served as an integration platform for all these elements. By using a common design parameter/performance metric/constraint interface this approach enabled us to rapidly find feasible design alternatives and drive the exploration process towards the Pareto-front by using a pruning and pushing approach. This overall architecture and the individual tools enabled us to rapidly adapt to changing domain constraints and mission requirements, and our team consistently delivered highly competitive results - compared to both human-engineered solutions and to other performers' designs - in both challenge domains.