Coordination Of Multi-agent Systems Research Articles

An event-driven behavior trees extension to facilitate non-player multi-agent coordination in video games

In this paper, we extend behavior trees (BTs), a behavior creation method that is popular in the video game industry, with three new types of nodes that facilitate the design and implementation of non-player characters (NPCs) that need to coordinate with each other. We provide an implementation and a methodology to use the coordination nodes of our extension appropriately, and we show how to use them to develop an application scenario. In the last years, coordination in multi-agent systems has been a very active research field, both from theoretical and practical points of view. Something similar has happened with the development of new tools for the video game industry. Our approach contributes to both areas by providing a novel extension that facilitates the design and implementation of agents that need to coordinate with each other. In video games, agents or NPCs are—as their name implies—characters that are not controlled by the player but by the game through an algorithmic, predetermined, or responsive behavior, or a more sophisticated AI technique. Some video games require NPCs with dynamic, credible, and intelligently unpredictable behaviors to keep players engaged and immersed. Instead of endowing NPCs with very complex individual behaviors, a feasible way to improve their unpredictability in an intelligent and credible manner is allowing them to coordinate with each other. Since BTs focus on the creation of individual behaviors, coordinated behaviors nowadays tend to be achieved by hard-coding the coordination itself. However, that ad hoc solution partially drives away some of the benefits that popularized BTs: Being visually intuitive, scalable, and reusable. For this reason, we propose an extension to BTs that developers can use to coordinate NPCs without going against the development paradigm: creating complex behaviors by designing an intuitive tree structure.

From Few to More: Large-Scale Dynamic Multiagent Curriculum Learning

A lot of efforts have been devoted to investigating how agents can learn effectively and achieve coordination in multiagent systems. However, it is still challenging in large-scale multiagent settings due to the complex dynamics between the environment and agents and the explosion of state-action space. In this paper, we design a novel Dynamic Multiagent Curriculum Learning (DyMA-CL) to solve large-scale problems by starting from learning on a multiagent scenario with a small size and progressively increasing the number of agents. We propose three transfer mechanisms across curricula to accelerate the learning process. Moreover, due to the fact that the state dimension varies across curricula, and existing network structures cannot be applied in such a transfer setting since their network input sizes are fixed. Therefore, we design a novel network structure called Dynamic Agent-number Network (DyAN) to handle the dynamic size of the network input. Experimental results show that DyMA-CL using DyAN greatly improves the performance of large-scale multiagent learning compared with state-of-the-art deep reinforcement learning approaches. We also investigate the influence of three transfer mechanisms across curricula through extensive simulations.

Communication-Free Multi-Agent Coordination in an Unknown Environment

Abstract We propose a communication-free control protocol for the coordination of multi-agent systems of autonomous vehicles. We consider a maze solving problem in which two agents must search an unknown maze as fast and as securely as possible. We formulate this problem using a discrete-event systems framework and present a coordination strategy for two agents to search a simple maze that involves no communication, thereby eliminating any cyber-security threats. Our strategy involves two agents having reverse priority decision-making policies, while searching the maze with a depth-first search control policy. Under our definition of inefficiency, which is a function of the number of dead end states that are searched by both agents, we show that our method achieves the minimum inefficiency cost, and thus performs as well as any communication-based solution.

Input-to-state stability for nonlinear systems with stochastic impulses

In this study, the problem of input-to-state stability (ISS) is systematically investigated for nonlinear systems with stochastic impulses. First, the ISS problem is considered for systems, where impulsive strengths are assumed to be stochastic and impulsive intervals are confined by the average impulsive interval. Then, the ISS problem is analyzed for nonlinear systems with both stochastic impulsive intensity and density. Several different cases are considered, i.e., cases where the impulsive instants are satisfied by a renewal process, Poisson process, semi-Markov chain and Markov chain. An example of coordination of multi-agent systems is provided to illustrate the effectiveness of the proposed new stability criteria.

Extortion Boosts Cooperation through Redistributing Strategies in Assortative Networked Systems

Networked evolutionary game theory is widely used to study the evolutionary dynamics in complex systems. In this work, we mainly investigate the influence of degree correlations of networks on the evolution of three memory-one strategies, i.e., unconditional cooperation, unconditional defection, and extortion. It is found that compared with the degree-uncorrelated network, assortative networks may inhibit the evolution of cooperation. The more assortative the network is, the less cooperation remains eventually. However, through micro dynamics analysis we reveal that by introducing extortion strategies, cooperation and defection are redistributed on the network, resulting in that cooperators can occupy the hub nodes while defectors survive among small-degree nodes. This work may provide potential clues to the coordination in multi-agent systems.

Affine Formation Algorithms and Implementation Based on Triple-Integrator Dynamics

This paper presents novel affine formation algorithms and implementations in different scenarios for the coordination of multi-agent systems with triple-integrator agent dynamics for both sampled-data and continuous-time settings. The agents in affine maneuver control are to be capable of producing required geometric shapes and simultaneously accomplishing desired maneuvers such as shearing, rotation, translation and scaling. From existing work, these tasks can be accomplished for systems whose agent dynamics are described using double-integrators and the agents communicate continuously in time. In some practical situations, however, the inter-agent communication may be limited to periodic intervals of time. Furthermore, a wide range of systems is governed by complex dynamics described with higher-orders. This paper presents two novel algorithms based on triple-integrator agent dynamics. Four implementation cases comprising of two scenarios each studied in both continuous-time and sampled-data cases are considered. Under the proposed algorithms, the collection of agents are capable of tracking time-varying targets which are affine transforms of the reference formation, if the leaders have knowledge of the required formation maneuvers. Detailed implementation results are presented to demonstrate the efficacy of the proposed algorithms.

Event-triggered distributed predictive control for asynchronous coordination of multi-agent systems

This paper investigates the event-triggered distributed predictive control (DPC) problem for multi-agent systems subject to bounded disturbances. A novel event-triggering mechanism which involves the neighbours’ information is derived for each agent to achieve a trade-off between resource usage and control performance. In such a framework, the DPC optimization problem is solved and information is exchanged only at triggering instants, thus achieving asynchronous coordination. To lower computation and communication consumption more significantly, a dynamic variable considering effects of neighbours is introduced to design a dynamic event-triggering condition and we show that larger inter-execution time can be obtained using the dynamic triggering mechanism. The theoretical conditions on ensuring feasibility and closed-loop stability are developed for these two triggering mechanisms, respectively. Finally, numerical simulations are given to illustrate the effectiveness of the proposed control strategy.

Semi-global leader-following coordination of multi-agent systems with input saturation and aperiodic intermittent communications

This paper addresses the semi-global leader-following coordination problem of general linear multi-agent systems, in which the control input of each agent is steered by aperiodically intermittent saturated actuator. Both the case with only one virtual leader and the case with multiple virtual leaders are discussed. By using multiple Lyapunov stability theory and applying algebraic Riccati equation-based low-gain feedback technique, sufficient conditions guaranteeing semi-global consensus tracking and semi-global containment tracking are provided. Numerical simulations finally verify the theoretical analysis.

Social or Individual Learning? An Aggregated Solution for Coordination in Multiagent Systems

There are mainly two different ways of learning for animals and humans: trying on yourself through interactions or imitating/copying others through communication/observation. How these two learning strategies differ and what roles they are playing in achieving coordination among individuals are two challenging problems for researchers from various disciplines. In multiagent systems, most existing work simply focuses on individual learning for achieving coordination among agents. The social learning perspective has been largely neglected. Against this background, this article contributes by proposing an integrated solution to decision making between social learning and individual learning in multiagent systems. Two integration modes have been proposed that enable agents to choose in between these two learning strategies, either in a fixed or in an adaptive manner. Experimental evaluations have shown that these two kinds of leaning strategies have different roles in maintaining efficient coordination among agents. These differences can reveal some significant insights into the manipulation and control of agent behaviors in multiagent systems, and also shed light on understanding the social factors in shaping coordinated behaviors in humans and animals.

Services Composition Modeling for Migrating Users in Intelligent Spaces

This research paper addresses the problem of modeling service between heterogeneous intelligent spaces, essentially to share complex services. In the last decade, development-focused health care systems in intelligent environments have taken great interest, especially through the development of systems that have proven useful for medical monitoring of users. However, these approaches are limited to small geographic areas such as a home or hospital. Current methods in cooperative coordination of multi-agent systems have investigated the problem of composing services, but only among the agents of a single society, leaving open the question of composition of services between different societies of agents. The purpose of this research is the development of a model service composition that can be used in the migration of users between smart spaces.

Motion planning and coordination of multi-agent systems

Development of a navigation scheme for a group of cooperative mobile robots is attempted in this paper. Potential field method is used to generate collision-free movement of a robot while it encounters other robots as obstacles. Finally, a cooperation scheme is proposed based on human behaviour to optimise motion strategies of robots treating each other as their obstacles. Computer simulations have been conducted for three different cases with four, eight and 12 robots negotiating a common dynamic environment. In each case, 100 different scenarios are considered and travelling times of all the robots are computed separately. The scenes in which the robots meeting collision is treated as a failure scenario and travelling time is not calculated. Performance of cooperation scheme has improved with the increase in a number of robots.

Motion planning and coordination of multi-agent systems

Motion planning and coordination of multi-agent systems

Recent Advances in Consensus of Multi-Agent Systems: A Brief Survey

In this paper, we mainly review the topics in consensus and coordination of multi-agent systems, which have received a tremendous surge of interest and progressed rapidly in the past few years. Focusing on different kinds of constraints on the controller and the self-dynamics of each individual agent, as well as the coordination schemes, we categorize the recent results into the following directions: consensus with constraints, event-based consensus, consensus over signed networks, and consensus of heterogeneous agents. We also review some applications of the very well developed consensus algorithms to the topics such as economic dispatch problem in smart grid and k -means clustering algorithms.

Coordination of multi-agent systems on interacting physical and communication topologies

A new framework is given for coordination of multi-agent systems that are interconnected by a physical coupling digraph. The edges of this graph represent physical couplings between agents that are fixed due to dynamical interactions. On top of the physical graph, distributed control protocols are designed where the allowed communications between agents for control purposes are prescribed by a second fixed communication digraph. The physical and communication digraphs are generally different and the combination of these two graphs forms a cyber-physical system. The interactions between physical and communication graphs are the focus of this paper. We consider different interactions between two graphs, including the case when their pinned Laplacian commutes, the case of the communication graph with diagonalizable pinned Laplacian, and the case of two general graphs. Moreover, within each graph, the relations between the agents can be either collaborative or antagonistic. To capture this, the theory of bipartite consensus is used. Coordination protocols for different cases are designed that are distributed with respect to the communication graph, and overcome the detrimental effects of the signed physical graph. The proposed control methods are illustrated by simulation examples.

Partial component consensus of leader-following multi-agent systems

Consensus problems, as basic topics in distributed coordination of multi-agent systems, have drawn a great deal of attention from different research fields. Generally, consensus refers to the asymptotic convergence of state variables of all agents with time evolution. In this paper, a concept on partial component consensus in multi-agent system is first given, which is a weaker dynamic behavior of group than the consensus in general, and then the problem of partial component consensus in leader-following first-order multi-agent system with the directed network topology is discussed. By designing an appropriate pinning control protocol and building corresponding error system, partial component consensus in multi-agent system is transformed into the partial variable stability of the error system. Using matrix theory and stability theory, a sufficient condition is given to realize partial component consensus in multi-agent system. Numerical simulations are given to illustrate the theoretical results.

A Survey on Fault Tolerant Multi Agent System

A mult i-agent system (MAS) is formed by a number of agents connected together to achieve the desired goals specified by the design. Usually in a mu lti agent system, agents work on behalf of a user to accomplish given goals. In MAS co-ordination, co-operation, negotiation and communication are important aspects to achieve fault tolerance in MAS. The mu lti-agent system is likely to fail in a distributed environment and as an outcome o f such, the resources for MAS may not be available due to the failure of an agent, mach ine crashes, process failure, software failure, communicat ion failure and/or hardware failure. Therefore, many researchers have proposed fault tolerance approaches to overcome the failure in MAS. So we have surveyed these approaches in this paper, whereby our contribution is threefold. Firstly, we have provided taxono my of faults and techniques in MAS. Secondly, we have provided a qualitative comparison of existing fault tolerance approaches. Thirdly, we have provided an evaluation of existing fau lt tolerance techniques. Results show that most of the existing schemes are not very efficient, due to various reasons like high computation costs, costly replication and large co mmunicat ion overheads.

Multifrequency-Coupled Oscillators for Distributed Multiagent Coordination

The coordination of multiagent systems in real environments receives considerable attention from research and industry. The design of coordination mechanisms should take into account the nature of the environment where the system is embedded. In this paper, the multiagent system is in an environment that features periodic properties. This environment is approached from a signal processing point of view to extract such properties. The coordination is performed by the proposed multifrequency-coupled oscillators (MuFCO) algorithm. It addresses the coordination of the multiagent system as a distributed collective synchronization mechanism. An operation example of MuFCO algorithm is shown, where it is used to coordinate consumptions in a smart grid. Thanks to the MuFCO algorithm, the multiagent system can be used to smooth the aggregated consumption of an electrical grid in a distributed way.

Coordination of multi-agent systems under switching topologies via disturbance observer-based approach

ABSTRACTIn this paper, a leader-following coordination problem of heterogeneous multi-agent systems is considered under switching topologies where each agent is subject to some local (unbounded) disturbances. While these unknown disturbances may disrupt the performance of agents, a disturbance observer-based approach is employed to estimate and reject them. Varying communication topologies are also taken into consideration, and their byproduct difficulties are overcome by using common Lyapunov function techniques. According to the available information in difference cases, two disturbance observer-based protocols are proposed to solve this problem. Their effectiveness is verified by simulations.

Research on Transaction Security Mechanism of Mobile Commerce in Mobile Internet based on MAS

In Mobile Internet (MI), existing uncertain factors, such as randomness, fuzziness and unpredictability, brought many security issues to on-line transactions. In on-line transactions of the mobile commerce based on mobile Internet, the trust is the premise and key which makes transactions smoothly.. Trust measurement between entities involves in trading volume, transaction time, personal income of consumer entity and its risk attitude to trust, and so on, which is difficult to be given accurately quantitative calculation. To find out the essential features of this kind of trust relationship, combined with cognitive theory and methods of social network in real life, in the research background of mobile commerce in Mobile Internet, by the research on problems in MI, such as trust, influencing factors to trust and trust mechanism, based on multi-Agent system coordination theory, a dynamic trust calculation model based on mobile Agent is given. The model can achieve the qualitative and quantitative conversion of trust. In order to effectively prevent credit speculation and fraud of malicious user, the paper presents evaluation methods of special attributes and punishment methods of. trust. To implement incentive mechanism for achieving credibility, the paper defines the timesensitive functions, which makes dynamic attenuation of the trust. The existing trust assignment methods of new users are improved, and a new method is proposed which sets dynamically initial trust of new users based on minimum trust of the previous system, and effectively resists moral hazard of discarding reputation information. Through leading the evaluation system of trading and weighting system into the multi-factor mechanism, the model reflects better the impact of subjective factors on trust calculation, such as individual preferences, risk attitudes, enhances the sensitivity of the trust algorithm on single property of transaction. The detailed theoretical analysis and simulation results show that the mechanism can effectively reveal and resolve the trust computing problem of mobile network transactions and provides a valuable new idea to transaction security in MI.

Hybrid subgroup coordination of multi-agent systems via nonidentical information exchange

Multiple coordination corresponds to a diversified synchronous regime where different subgroups in one multi-agent system may emerge several distinct collective behaviors respectively. This paper is devoted to propose and analyze hybrid subgroup coordination of multi-agent systems under generic directed topologies. Due to unanticipated practical situations, agents in one subgroup are supposed to achieve a synchronous scenario asymptotically, such as average of agents’ initial states or its bounded region, while there is no agreement between any two distinct subgroups. Certain assumptions are imposed on the underlying topology for agents to evolve into several subgroups. Explicit distributed control algorithms are designed by adopting nonidentical extra-subgroup and intra-subgroup information exchange, and detailed convergence analysis is presented based on techniques of graph and matrix. Relevant simulation work is further carried out to illuminate the developed theoretical results.