Federated learning is expected to be applied to many problems in control, given the increasing interest in data-driven control. We propose two federated system identification schemes to estimate parameters using observational data from multiple clients without transmitting the data. When performing federated learning, scheduling becomes necessary due to the system's instability and communication. Therefore, we also propose a fast scheduling method with its theoretical performance guarantee. We conduct experiments on federated recursive system identification using several scheduling algorithms, and investigates the accuracy and convergence speed.

In recent years, reinforcement learning has advanced in various fields such as autonomous driving, behavior analysis in power markets, and robot control. These studies have demonstrated the utility of applying MAS (Multi-Agent System), where multiple agents cooperate and compete with each other. In the learning process within MAS, a significant challenge is that irregular fluctuations in the environment caused by the actions of other agents increase environmental uncertainty and destabilize learning. To overcome this challenge, research has been reported on methods to improve learning efficiency by effectively utilizing the experience data of other agents. Previous studies have reported cases where sharing all data improved the learning efficiency of agents in cooperative or competitive relationships, as well as cases where sharing only a portion of the experience data improved learning efficiency. This paper aims to improve learning efficiency in MAS by emphasizing the similarity between agents as a criterion for data sharing and demonstrating its effectiveness. Specifically, we propose a new method that selects shared data based on the similarity between agents and treats it as experience data that complements one's own experience. Furthermore, we demonstrate that the proposed method improves the learning efficiency of agents through simulations involving asymmetric agents.

This note addresses the conversion problem of a priori designed unstructured controllers to observer-structured controllers. Observer-structured controllers have been proposed as a structural extension of Luenberger observer-based controllers to accommodate linear parametric plant systems with unspecified nominal state-space matrices, and the usefulness has been demonstrated in the literature. However, the conversion process in the literature uses only the mathematical models of the plant systems and the a priori designed unstructured controllers, which might degrade the plant state estimation performance of the converted observer-structured controllers due to the possibility of unexpected disturbances, unmodeled dynamics, etc. of the practical operating environment. Therefore, a conversion method which well reflects the practical operating environment has been desired. To this end, this note proposes a conversion method, which is based on the least square method, by using the data of the plant systems controlled by the a priori designed unstructured controllers under practical operating environment, and illustrates the effectiveness of our method.