A new achievement by researchers at the College of Engineering of the University of Tehran in developing the next generation of intelligent and autonomous systems
According to the Public Relations Office of the College of Engineering, the article titled “A Distributed Primal-Dual Method for Constrained Multi-Agent Reinforcement Learning With General Parameterization,” authored by Ali Kaheh, the top-ranked graduate of the M.Sc. program in Electrical Engineering at the University of Tehran, and Dr. Hamed Kebriaei, a faculty member of the School of Electrical and Computer Engineering within the College of Engineering, was published in 2026 in the prestigious journal IEEE Transactions on Automatic Control. This journal, with a scientific impact index of SJR = 3.929, has ranked first among 382 international journals in the field of systems and control in the latest SCImago ranking.
This research presents a novel method for solving constrained multi-agent reinforcement learning problems. In many intelligent systems, a set of agents must make decisions simultaneously and independently while also achieving shared objectives. Most existing methods for training such agents require a central controller or centralized coordination, which limits their implementation in real-world, large-scale environments. In this study, researchers from the University of Tehran propose a fully distributed method in which each agent can learn and make decisions using only its local information and that of its neighboring agents.
The proposed approach is based on the Primal-Dual algorithm and an Actor-Critic structure, enabling intelligent agents to perform online and independent learning while satisfying shared system constraints. The researchers also demonstrate that the dual variables, or Lagrange multipliers, converge to a consensus among agents, and that the proposed algorithm theoretically converges to an equilibrium point.
One of the key features of this research is its ability to employ general parameterized policies in environments with large state spaces, increasing its applicability to complex real-world problems. In addition, the performance of the algorithm has been evaluated using a cooperative and dynamic model of the Cournot economic game, and the results confirm the effectiveness of the proposed method under complex and stochastic conditions.
Multi-agent reinforcement learning is considered one of the most important branches of artificial intelligence, with wide-ranging applications in fields such as robotics, smart grid control, urban traffic management, autonomous systems, and economic market modeling. The development of distributed and scalable methods in this area can play a significant role in advancing the next generation of intelligent and autonomous systems.
This article is available at the following link:
Your Comment :