A new, $8.7 million project—Design. R–AI-assisted CPS Design—involves pathbreaking work for the Defense Advanced Research Projects Agency as future cyber-physical systems will rely less on human control and more machine learning algorithms and artificial intelligence processors.
Smart grid, driver-assist and autonomous automobile systems, health and biomedical monitoring, smart cities, robotics systems, and new agricultural technologies are just a few CPS that interact with users in a lot of ways that change with context.
“Our vision is the reformulation of the conventional engineering process of CPS as a continuously learning, self- improving process of collaborative discovery,” said Péter Völgyesi, principal investigator and senior research scientist in the Vanderbilt Institute for Software Integrated Systems.
Data-driven AI methods will play an increasing role in the design and implementation of CPS and are expected to have a significant impact on engineering processes. Advances in deep learning and reinforcement learning, plus vast increases in computing power offer the team the potential to develop and deliver fundamentally new design capabilities.
The goal is to develop open-source AI-based co-designers that are integrable with CPS design flows and tool suites. The critical challenge is the reformulation of notoriously hard problems in model-based CPS design as AI/machine learning problems and integrating the solution into continuous learning, AI-based co-designers that interact with human designers. The result could have a profound effect on design productivity and agility in the CPS design flow process, Völgyesi said.
“Our Vanderbilt team also has gained a broad and deep understanding of the fundamental and practical limitations of model- and component-based design of complex CPS as prime developers of the OpenMETA design automation tool suite for DARPA,” Völgyesi said. For more than 20 years, ISIS has pioneered generations of metaprogrammable tool suites for modeling and model transformation.
“We have identified and understand the roadblocks and looked at existing and emerging AI methods to solve those problems,” he said.
The chief scientist on the four-year project is Csaba Szepesvári, University of Alberta computer science professor and Foundations leader at Google DeepMind. Miklós Maróti, Ph.D.’02, a mathematics professor at University of Szeged, Hungary, is a key team member along with several Vanderbilt engineering faculty.