Earticle

In this study, we analyze the dynamic characteristics of the Cucker-Smale model with heterogeneous free will (CSF). Most existing research has examined flocking when external perturbations converge to identical values; however, we investigate the case in which agents have different free-will vectors. We first prove that a positive lower bound exists for velocity differences between agents with heterogeneous free will. This implies that flocking cannot occur in the CSF. In particular, in the case of a nontrivial initial velocity and free will, we demonstrate that the distance between agents increases linearly with time. We also obtain an explicit upper bound for the maximum distance between agents when the communication rate is greater than zero and less than 1/2. The maximum distance growth rate depends on the communication rate. Finally, we validate the theoretical results through numerical simulations.

In order to ensure second-order multi-agent systems (MAS) realizing consensus more quickly in a limited time, a new protocol is proposed. In this new protocol, the gradient algorithm of the overall cost function is introduced in the original protocol to enhance the connection between adjacent agents and improve the moving speed of each agent in the MAS. Utilizing Lyapunov stability theory, graph theory and homogeneity theory, sufficient conditions and detailed proof for achieving a finite-time consensus of the MAS are given. Finally, MAS with three following agents and one leading agent is simulated. Moreover, the simulation results indicated that this new protocol could make the system more stable, more robust and convergence faster when compared with other protocols.

This study investigates the problem of leader-following consensus for multi-agent systems with output saturations. This study assumes that the agents are described as a neutrally stable system, and the leader agent generates the bounded trajectory within the saturation level. Then, the objective of the leader-following consensus is to track the trajectory of the leader by exchanging information with neighbors. To solve this problem, we propose an observer-based distributed consensus algorithm. Then, we provide a consensus analysis by applying the Lyapunov stability theorem and LaSalle's invariance principle. The result shows that the agents achieve the leader-following consensus in a global sense. Moreover, we can achieve the consensus by choosing any positive control gain. Finally, we perform a numerical simulation to demonstrate the validity of the proposed algorithm.

The growing complexity of 5G/6G, IoT, and edge networks exposes the limits of single-agent LLM control in scalability, real-time performance, and reliability. Multi-Agent Systems (MAS) combined with LLMs offer a path forward. This paper surveys state-of-the-art LLM-based MAS for network management. We propose a taxonomy that re-interprets the Centralized/Decentralized/Hybrid split under two networking-specific axes (a strict LLM-core inclusion criterion and an anchoring on the NetOps lifecycle), complemented by a role-based lens (Coordinator, Translator, Negotiator). We finally identify key challenges such as hallucination and inference latency, and outline directions toward trustworthy autonomous networks.

This study investigates the emergence of emotional dynamics and intrinsic motivation within multi-agent reinforcement learning (MARL) systems. Traditional MARL frameworks rely solely on extrinsic task rewards, which often limit exploration and adaptability. To address this, we propose an Affective-Motivated MARL (AM-MARL) framework where agents integrate curiosity-based intrinsic rewards and emotionmodulated affective feedback alongside extrinsic reinforcement. Agents operate in a continuous multiagent environment, learning through Q-learning, Actor- Critic, or Advantage Actor-Critic (A2C) methods depending on their action space. The intrinsic reward is defined as the state-prediction error between observed and expected future states, while the affective reward arises from temporal changes in emotional state and the social influence among peers. Experimental results show that incorporating intrinsic and affective rewards enhances exploration coverage, stabilizes emotional trajectories, and improves coordination efficiency compared to extrinsic-only baselines. These findings suggest that emotional feedback, when coupled with curiosity-driven intrinsic signals, fosters more humanlike adaptability, cooperative intelligence, and stable affect regulation in MARL environments.

최근 artificial intelligence (AI) 에이전트가 다양한 분야에서 활용되면서 여러 에이전트들이 협력하여 작업을 수행하는 다중 에이전트 시스템이 널리 활용되고 있다. 다중 에이전트 시스템은 여러 에이전트가 작업을 분담하고 협력하여 수행함으로써 높은 확장성과 성능 향상을 기대할 수 있으나, 종래의 연구들은 에이전트 간 협력 구조나 성능 향상에 초점을 맞추고 있고, 에이전트 수 증가에 따른 성능 대비 자원 효율성에 대한 분석은 부족하다는 한계가 있다. 이에 본 연구에서는 에이전트 수 변화에 따른 다중 에이전트 시스템의 처리 성능과 효율 변화를 정량적으로 분석하였다. 실험 결과, 에이전트 수 증가에 따라 처리 시간은 평균 50.6% 감소하였지만, 데이터 처리량은 에이전트 수 증가에 따라 증가하다가 일정 수준에서 포화하는 경향을 보였다. 에이전트 수에 따른 데이터 처리량 대비 비용을 고려한 효율성은 에이전트 수가 증가함에 따라 초기 구간에서는 평균 19.7 % 향상되었지만 점차 감소하였으며, 이를 통해 작업 크기에 따른 최적의 에이전트 수를 도출할 수 있음을 확인하였다.

With the recent widespread adoption of artificial intelligence (AI) agents across various fields, multi-agent systems-in which multiple agents collaborate to perform tasks-are becoming increasingly prevalent. While multi-agent systems offer the potential for high scalability and improved performance by allowing agents to divide and collaborate on tasks, they face the limitation of a lack of analysis regarding resource efficiency relative to performance as the number of agents increases. Previous studies have primarily focused on the cooperative structure among agents or performance improvements, and there is a lack of quantitative analysis regarding the decline in efficiency resulting from an increase in the number of agents. Therefore, this study quantitatively analyzed the changes in processing performance and efficiency of multi-agent systems as the number of agents varies. The experimental results showed that as the number of agents increased, processing time decreased by an average of 50.6%, while data throughput initially increased with the number of agents but tended to plateau at a certain level. Efficiency improved by an average of 19.7% in the initial phase before declining, confirming that an optimal number of agents exists depending on the task scale.

다수의 자동 운반 차량(AGV)을 운용하는 시스템에서는 AGV 간의 충돌, 교착 상태 발생, 그리 고 비효율적인 대기 시간 증가로 인해 전체 시스템의 효율성이 저하되는 문제가 빈번하게 발생한 다. 본 논문은 이러한 문제를 해결하기 위해, 연속적인 공간으로 모델링된 공장 환경에서 100대의 AGV가 상호 협력을 통해 최적의 경로를 계획하도록 하는 PPO(Proximal Policy Optimization) 기 반 분산 강화학습 모델의 실험 설계를 제안한다. 제안하는 설계는 시뮬레이션 환경 구성, 개별 AGV 에이전트 모델링, 보상 함수 체계, 그리고 학습 구조를 체계적으로 정의함으로써, AGV 간의 충돌을 회피하고 교착 상태를 방지하며 이동 경로상의 비효율을 최소화하는 것을 목표로 한다. 이 설계를 통해 대규모 다중 AGV 시스템의 운영 안전성과 효율성을 향상시킬 수 있을 것으로 기대한다.

Dynamic adaptation has become an important issue when designing and developing distributed applications in order to manage their Quality of Service (QoS). This is especially challenging when distributed applications run in environments in which resources vary unpredictably over time. To deal with fluctuations in resource availability and inherent heterogeneity of distributed environments requires dynamic adaptation of each application. This trend motivates the design of resource-aware applications ensuring a given performance level by adapting their behavior to changing contexts. To tune the use of resources, adaptive systems include the necessary mechanisms to modify applications' behavior. This paper presents a general distributed middleware for enabling behavior adaptation of distributed applications. The middleware combines application designer specification of alternative execution behaviors with information about the execution environment context for deciding when and how to adapt itself according to the available resources. A description language specifies applications' behavior and the description of their related resource use. The QoS management requires a common execution model for all applications. Simulations of the QoS management of heterogeneous applications illustrate our proposal and show the benefits. In addition, we discuss lessons learned from our experience.

How to generate the task-oriented optimal agent coalition is a key issue of multi-agent system, which is a typical optimization problem. In this paper, an improved particle swarm optimization (IPSO) is proposed to solve this problem. In order to overcome the premature and local optimization problem in traditional particle swarm optimization (PSO), we proposed a variation of inertia weight PSO algorithm by analyzing the feasibility of particle optimization process in PSO. Compared with several well-known algorithms such as PSO, ACO, experimental results show that the global search capability of IPSO has been significantly improved and IPSO can effectively avoid premature convergence problem. Also it can solve the multi-agent coalition formation problem effectively and efficiently.

Multi-agent systems fit nicely into domains that are naturally distributed and require artificial intelligence technology. The autonomous network-based information services integration architecture has been designed to satisfy the multiservice utilization in rapidly changing environments. However, the increase in the total number of user requests and changes lead to the unbalancing load in the system and the overload in the locality. This paper proposes a new strategy to solve such problem. Autonomous load distribution can be achieved through the integrated access method, which reduces the total load of the system for the number of Pull-MAs sent to the system decrease. In addition, the information structure of integrated service area is effective to improve the ratio of the satisfaction of Pull-MAs with joint request on one node. As a result, the homogeneous distribution of the separated services requests and correlated services requests is guaranteed autonomously. The simulation witnesses the success of the proposed mechanism.

This paper presents RE4Gaia, which is a requirements modeling approach for the development of multi-agent systems that extends the Gaia methodology. The approach focuses on dealing with the organizational structure as a means to adequately capture and understand required roles and associated functions in the context of an organization prior to the analysis and design of MAS using Gaia. In addition, a traceability framework is introduced to facilitate moving from the requirements models to the analysis and design models proposed in Gaia.

Aimed at the deficiency of existing modeling method of traffic guidance system based on Multi-Agent, the advantage of Unified Modeling Language (UML) and Hierarchical Colored Petri Nets (HCPN) is combined to model the system. The modeling method using UML and HCPN is first put forward, and more UML diagrams are constructed so as to model the framework of traffic guidance systems. Moreover, the mapping rules from UML model to HCPN is set up, the formal model of traffic guidance system is gotten, and design error may be found by formal verification and validation of HCPN model. So UML model can be improved and correct UML model can be gotten.

Distributed renewable energy generation is lack of intelligent information processing and decision-making section. The above problems can be solved effectively by the means of intelligent information processing technology based on the dynamic structure of multi-agent for distributed renewable energy. In order to solve the problems of less data and poor information, this paper puts forward fuzzy hyperbolic model and attempts to use dynamic intelligent information processing technology based on multi-agent to change the unreliable, inaccurate information into be full, reliable and accurate. And then the information we have obtained could be intelligent fitted, filtered and decided. The self-organizing self-learning and reasoning ability of multi-agent are considered when the information processing. The intelligent information processing section is comprised of three layers and two kinds of agents. The two kinds of agents are Bus Agents and Coordination Agents, and the three structural layers are the data layer, the filter layer and the policy making layer. Moreover the information have been processed could be utilized.

Physicians, in order to study the causes of cancer, detect cancer earlier, prevent or determine the effectiveness of treatment, and specify the reasons for the treatment ineffectiveness, need to access accurate, comprehensive, and timely cancer data. The cancer care environment has become more complex because of the need for coordination and communication among health care professionals with different skills in a variety of roles and the existence of large amounts of data with various formats. The goals of health care systems in such a complex environment are correct health data management, providing appropriate information needs of users to enhance the integrity and quality of health care, timely access to accurate information and reducing medical errors. These roles in new systems with use of agents efficiently perform well. Because of the potential capability of agent systems to solve complex and dynamic health problems, health care system, in order to gain full advantage of E- health, steps must be taken to make use of this technology. Multi-agent systems have effective roles in health service quality improvement especially in telemedicine, emergency situations and management of chronic diseases such as cancer. In the design and implementation of agent based systems, planning items such as information confidentiality and privacy, architecture, communication standards, ethical and legal aspects, identification opportunities and barriers should be considered. It should be noted that usage of agent systems only with a technical view is associated with many problems such as lack of user acceptance. The aim of this commentary is to survey applications, opportunities and barriers of this new artificial intelligence tool for cancer care information as an approach to improve cancer care management.

This paper presents consensus algorithms by integrating cooperative control and adaptive control laws for multi-agent systems with unknown nonlinear uncertainties. An ideal multi-agent system without uncertainties is introduced first. The cooperative control law, based on an artificial potential function, is designed to make the ideal multi-agent system achieve consensus under a fixed and connected undirected graph. The presence of uncertainties will degenerate the performance, or even destabilize the whole multi-agent system. The $L_1$ adaptive control law is therefore introduced to handle unknown nonlinear uncertainties. Two different consensus cases are considered: 1) normal consensus-where all agents reach an agreement on an initially undetermined position and velocity, and 2) consensus with a virtual leader-where all agents' states converge to the virtual leader's states. Under a fixed and connected undirected graph, the presented consensus algorithms enable the real multi-agent system to stay close to the ideal multi-agent system which achieves consensus with or without a virtual leader. Simulation results of 2-D consensus with nonlinear uncertainties are provided to demonstrate the presented algorithms.

In this paper, the formation control problem is investigated for discrete-time multi-agent systems with unknown nonlinear dynamics by means of the iterative learning approach. For networks with switching topology, a distributed iterative learning scheme is developed using the local formation error data with anticipation in time, and a sufficient condition is derived to guarantee that the desired formation can be preserved during the whole finite-time motion or operation process, even in the presence of initial formation errors. Simulation results illustrate the effectiveness of the proposed method.

This paper discusses a design methodology of cooperative path planning for dynamical multi-agent systems with spatial and temporal constraints. The cooperative behavior of the multi-agent systems is specified in terms of the objective function in an optimization formulation. The path of achieving cooperative tasks is then generated by the optimization formulation constructed based on a differential flatness approach. Three scenarios of multi-agent tasking are proposed at the cooperative task planning framework. Given agent dynamics, both spatial and temporal constraints are considered in the path planning. The path planning algorithm first finds trajectory curves in a lower-dimensional space and then parameterizes the curves by a set of B-spline representations. The coefficients of the B-spline curves are further solved by a sequential quadratic programming solver to achieve the optimization objective and satisfy these constraints. Finally, several illustrative examples of cooperative path/task planning are presented.

In this paper, the consensus problem of second-order nonlinear multi-agent systems with directed communication topology is discussed. Information for the states of neighbour agents is quantized using a uniform quantizer. The sliding mode control law is designed based on quantized state information and the stability analyzed by using Lyapunov theory. Additionally, an iterative learning control law based on sliding mode errors is developed. Thereby, a novel consensus control protocol consisting of a sliding mode control law and an iterative learning approach is achieved, where the purpose of applying the sliding mode control law is to eliminate non-repeatable uncertainties and the iterative learning control law is to remove repeatable disturbances. Also, the convergence of the proposed control protocol is analyzed in the frequency domain. Finally, two cases are provided to illustrate the effectiveness of theoretical analysis.

This paper is devoted to the stochastic bounded consensus tracking problems of second-order multi-agent systems, where the control input of each agent can only use the information measured at the sampling instants from its neighbors or the virtual leader with a time-varying reference state, the measurements are corrupted by random noises, and the signal sampling process induces the small sampling delay. The augmented matrix method, the probability limit theory and some other techniques are employed to derive the necessary and sufficient conditions guaranteeing the mean square bounded consensus tracking. We show that the convergence of the proposed protocol simultaneously depends on the constant feedback gains, the network topology, the sampled period and the sampling delay, and that the static consensus tracking error depends on not only the above mentioned factors, but also the noise intensity and the upper bound of the velocity and the acceleration of the virtual leader. The obtained results cover no sampling delay as its one special case. Simulations are provided to demonstrate the effectiveness of the theoretical results.