A new AI framework improves management of complex systems with unequal decision-makers, like smart grids, traffic networks, and autonomous vehicles.

A new artificial intelligence breakthrough developed by researchers in the College of Engineering and Computer Science at Florida Atlantic University offers a smarter, more efficient way to manage complex systems that rely on multiple decision-makers operating at different levels of authority.

This novel framework, recently published in IEEE Transactions on Systems, Man and Cybernetics: Systems, could significantly impact the future of smart energy grids, traffic networks and autonomous vehicle systems – technologies that are becoming increasingly central to daily life.

In many real-world systems, decisions don’t happen simultaneously or equally. A utility company might decide when to cut power during peak hours, and households must adjust their energy use in response. In traffic systems, central controllers dictate signals while vehicles adapt accordingly.

“These types of systems operate under a power hierarchy, where one player makes the first move and others must follow, and they’re more complicated than typical AI models assume,” said Zhen Ni, Ph.D., senior author, IEEE senior member and an associate professor in the Department of Electrical Engineering and Computer Science. “Traditional AI methods often treat every decision-maker as equal, operating at the same time with the same level of influence. While this makes for clean simulations, it doesn’t reflect how decisions are actually made in real-world scenarios – especially in environments full of uncertainty, limited bandwidth and uneven access to information.”

To address this, Ni and Xiangnan Zhong, Ph.D., first author, IEEE member and an associate professor in the Department of Electrical Engineering and Computer Science, designed a new AI framework based on reinforcement learning, a technique that allows intelligent agents to learn from interacting with their environment over time.

Their approach adds two key innovations. First, it structures the decision-making process using a game theory model called the Stackelberg-Nash game, where a “leader” agent acts first and “follower” agents respond in an optimal way. This hierarchy better mirrors systems like energy management, connected transportation and autonomous driving. Second, the researchers introduced an event-triggered mechanism that reduces the computational burden.

“Instead of constantly updating decisions at every time step, which is typical of many AI systems, our method updates decisions only when necessary, saving energy and processing power while maintaining performance and stability,” said Zhong.

The result is a system that not only handles the power asymmetry between decision-makers but also deals with mismatched uncertainties – cases where different players operate with varying levels of information and predictability. This is especially critical in environments like smart grids or traffic control systems, where conditions change rapidly and resources are often limited. The framework allows for a more robust, adaptive and scalable form of AI control that can make better use of limited bandwidth and computing resources.

“This work fills a crucial gap in the current AI landscape. By developing a method that reflects real-world decision hierarchies and adapts to imperfect information, Professors Zhong and Ni are helping us move closer to practical, intelligent systems that can handle the complexity of our modern infrastructure,” said Stella Batalama, Ph.D., dean of the College of Engineering and Computer Science. “The implications of this research are far-reaching. Whether it’s optimizing power consumption across cities or making autonomous systems more reliable, this kind of innovation is foundational to the future of intelligent technology. It represents a step forward not just for AI research, but for the everyday systems we depend on.”

Backed by rigorous theoretical analysis and validated through simulation studies, Zhong and Ni demonstrated that their event-triggered reinforcement learning method maintains system stability, ensures optimal strategy outcomes and effectively reduces unnecessary computation. The approach combines deep control theory with practical machine learning, offering a compelling path forward for intelligent control in asymmetric, uncertain environments. Two related journal articles have recently been published on IEEE Transactions on Artificial Intelligence as well. The research work is mainly supported by the National Science Foundation and the United States Department of Transportation.

The research team is now working on expanding their model for larger-scale testing in real-world scenarios. Their long-term vision is to integrate this AI framework into operational systems that power cities, manage traffic and coordinate fleets of autonomous machines – bringing the promise of smarter infrastructure one step closer to reality.

-FAU-