Evolutionary Computation as a Paradigm for Engineering Emergent Behaviour in Multi-Agent Systems

Optimization of performance in collective systems often requires altruism. Emergence and stabilization of altruistic behaviors are difficult because the agents incur a cost when behaving altruistically. In this paper we propose a biologically inspired strategy to learn stable altruistic behaviors in artificial multi-agent systems, namely reciprocal altruism. Our multi-agent system is made up of autonomous agents with a behavior-based architecture. Agents learn the most suitable cooperative strategy for different environments by means of a reinforcement learning algorithm. Each agent receives a reinforcement signal that only measures its individual performance. Simulation results show how the multi-agent system learns stable altruistic behaviors, so reaching optimal (or near-to-optimal) performances in unknown and changing environments.

Multi-agent systems exhibit unexpected, emergent behavior as a result of the complexity of agent behaviors and their interactions. Despite significant research interest in the past decades, computational methods to identify and analyze emergence as it happens are still needed. This paper proposes a software architecture for identifying emergent behavior in a multi-agent system as it happens, using interval-based snapshots and emergent behavior metrics. We propose various distance functions to compare between the multi-agent system under analysis and systems that have been previously shown to exhibit emergent behavior.

Implementing collective behaviour in cooperative multi-agent systems requires several practical constraints to be addressed. In some environments, communication bandwidth is a critical constraint which may compromise the intended cooperative behaviour. This paper introduces a bio-inspired model which invokes collective behaviour in a multi-agent system using passive sensing without any explicit inter-agent communication. An agent looks for the majority of its neighbours either in its left or its right half space using two sensors. For a source localization problem, we compare performance of the proposed model using passive sensing against the well known school-of-fish collective behaviour models using ideal explicit inter-agent communication. For different cue strengths and neighbourhood radii, our results show that the proposed strategy boosts higher levels of group cohesion to make up for the information loss and in certain conditions performs better than the other collective behaviour models.

Multiagent systems in which independent software agents interact with each other to achieve common goals, complete distributed tasks concurrently under autonomous control. Agent communication has been shown to be an important factor in coordinating efficient group behavior in agents. Most research on training or evolving group behavior in multiagent systems used predefined agent communication protocols. Designing agent communication becomes a complex problem in dynamic and large-scale systems. In order to solve this problem, in our previous research we proposed a method applying genetic programming techniques, in particular Automatically Defined Function Genetic Programming (ADF-GP) (K. Mackin and E. Tazaki, 1999), to allow agents to autonomously learn effective agent communication messaging. For this research we take this approach further and combine training of the agent behavior as well as the communication protocol. By training both behavior and communication we expect to further optimize the system performance. A software simulation of a multiagent transaction system is used to observe the effectiveness of the proposed method.

Multiagent systems, in which independent software agents interact with each other to achieve common goals, complete concurrent distributed tasks under autonomous control. Agent Communication has been shown to be an important factor in coordinating efficient group behavior in agents. Most researches on training or evolving group behavior in multiagent systems used predefined agent communication protocols. Designing agent communication becomes a complex problem in dynamic and large‐scale systems. In order to solve this problem, in this paper we propose a new application of existing training methods. By applying Genetic Programming techniques, namely Automatically Defined Function Genetic Programming (ADF‐GP), in combination with pheromone communication features, we allowed the agent system to autonomously learn effective agent communication messaging for coordinated group behavior. A software simulation of a multiagent transaction system aiming at e‐commerce usage will be used to observe the effectiveness of the proposed method in the targeted environment. Using the proposed method, automatic training of a compact and efficient agent communication protocol for the multiagent system was observed.

Welcome to the Eighth Annual Workshop on Evolutionary Computation and Multi-Agent Systems and Simulation (ECoMASS 2014)! Evolutionary computation (EC) and multi-agent systems and simulation (MASS) both involve populations of agents. EC is a learning technique by which populations of individual agents adapt according to the selection pressures exerted by an environment; MASS seeks to understand how to coordinate the actions of a population of (possibly selfish) autonomous agents that share an environment so that some outcome is achieved. Both EC and MASS have top-down and bottom up features. For example, some aspects of multi-agent system engineering (e.g., mechanism design) are concerned with how top-down structure can constrain or influence individual decisions. Similarly, most work in EC is concerned with how to engineer selective pressures to drive the evolution of individual behavior towards some desired goal. Multi-agent simulation (also called agent-based modeling) addresses the bottom-up issue of how collective behavior emerges from individual action. Likewise, the study of evolutionary dynamics within EC (for example in coevolution) often considers how population-level phenomena emerge from individual-level interactions. Thus, at a high level, we may view EC and MASS as examining and utilizing analogous processes. It is therefore natural to consider how knowledge gained within EC may be relevant to MASS, and vice versa; indeed, applications and techniques from one field have often made use of technologies and algorithms from the other field. Studying EC and MASS in combination is warranted and has the potential to contribute to both fields. The ECoMASS Workshop at GECCO has a successful history as a forum for exploring precisely this intersection, and we are looking forward to another year of stimulating discussion, bringing together experts as well as novices in both areas, to engage in dialogue about their work. This year's participants bring a variety of topics for discussion, including migratory flows, the beating of the heart, embedded systems, pelotons and cyclists, and flocking behavior. This year, we will also be adding a research slam. We will be inviting presenters who are presenting in the general conference to give us a quick presentation about their work. We hope that this will further encourage additional lively discussion about a wide range of topics. We also encourage presenters to demonstrate their software live during a breakout session, enabling additional opportunities for one-on-one dialogue and discussion of actual running systems. Through these additions to the workshop, we hope to continue to innovate, and develop the conversations around the interesting intersection of evolutionary computation and multi-agent systems and simulations!

In this paper, we propose a formal approach to study the emergent behaviors in multiagent systems with limited communications and environmental constraints. Temporal logic and formal grammars are employed to specify the formal model of multiagents, and based on that an automated approach can be used to identify the emergent behaviors due to interactions among agents. Instead of exploring all possible system state spaces, which may grow exponentially, the proposed approach computes the local group polarization and the local group momentum for generating the emergent behavior set, and accordingly from which the emergent behaviors can be detected. To facilitate simulation and analysis, a MATLAB GUI is further designed to enable us to flexibly select control parameters in formal models of multiagent systems. Simulation results are provided to verify the proposed approach.

This paper discusses a mechanism to ensure safe behaviors of HMSs (Holonic Manufacturing Systems). The change from low-variety high-volume to high-variety low-volume production requires highly flexible and adaptive manufacturing systems. Multi agent systems, in which decisions are made through cooperation among autonomous and cooperative elements, fulfill these requirements by exploiting full abilities of individual elements while eliminating various bottlenecks that exist in conventional systems. However, highly adaptive features induce non-deterministic behaviors of systems, and make it difficult to adopt multi agent systems as primary bases of manufacturing systems. In order to apply these systems to large and complicated applications, mechanisms that make their behaviors more predictable are essential. HMSs are a kind of multi agent systems, and an HMS safety ensuring mechanism proposed here is one of the attempts to make behaviors of multi agent systems more predictable.

Multiagent systems, in which independent software agents interact with each other to achieve common goals, complete concurrent distributed tasks under autonomous control. We propose a method applying genetic programming (GP) to allow agents to autonomously learn effective agent communication messaging in order to achieve group behavior. A software simulation of a multiagent system will be used to observe the effectiveness of the proposed method. Using the proposed method, emergence of a compact and efficient agent communication protocol for the multiagent system was observed.

Multi-Agent Systems (MAS) are extensively studied and defined differently in literature. A MAS can be described as an environment with multiple agents that interact with each other or the environment, following predetermined rules to achieve individual or shared goals. This work represents a survey focused on collaborative and non-collaborative behaviors within MAS. The survey presents the key characteristics, benefits and drawbacks of each behavior type and provides a structured review of the literature, emphasizing on the applications of these behaviors in different domains. Understanding these behaviors in MAS can have benefits in both academic research and day-to-day life. This understanding may help researchers and practitioners to design and implement MAS and to provide guidance for different behaviors that may be suitable based on specific scenarios.

Multi agent systems (MAS) play an important role in many of industrial software systems today. MAS are distributed systems and there is no central control for the agents in most of the systems. The lack of central control and the local view of each agent from the whole system may cause some unexpected behaviors which is called emergent behavior. Detecting and fixing these unexpected behaviors is more valuable and cost effective in the requirement and design phases rather than the implementation. In this paper, a method for the detection of emergent behaviors in MAS focusing on analysis phase of MaSE methodology and fixing them in the design phase is presented. The contributions of this work are investigating a specific type of emergent behaviors, the detection of them in MaSE methodology, and proposing solutions for the detected emergent behaviors. The details and the architecture of the system are presented. The method and how to apply this on MaSE is shown along with a case study followed by the discussion part.

As utility calculus cannot account for an important part of agents’ behaviour in Multi-Agent Systems, researchers have progressively adopted a more normative approach. Unfortunately, social laws have turned out to be too restrictive in real-life domains where autonomous agents’ activity cannot be completely specified in advance and the complexity of the system is ever changing. The idea of Rights is a halfway concept between anarchic and off-line constrained interaction. Rights improve coordination and facilitate social action in Multi-Agent domains through adjusting the coordination mechanisms to the complexity of the system. So far rights have not been tested or proven experimentally. We are comparing experimentally the three mentioned interaction architectures in the domain of agent-based traffic simulation.

Analysis of motor control and behavior in multi agent systems by means of artificial neural networks

The Petri nets (PN)-based analytical approach to describing the agent behaviour in MAS (multi agent systems) is presented. PN yield the possibility to express the MAS behaviour by means of the vector state equation in the form of linear discrete system. Thus, the modular approach to the creation of the MAS model can be successfully used too. Three different interconnections of modules (agents, interfaces, environment) expressed by PN subnets are introduced. Special attention is paid to conflict resolution and machine learning methods based on modelling tools, detection and resolution of conflicts. The approach makes possible to use methods of linear algebra. In addition, it can be successfully used at the system analysis (e.g. the reachability of states), at testing the system properties, and even at the system control synthesis.KeywordsAgentPetri NetNegotiationConflict ResolutionMachine Learning

Coalition formation provides a versatile framework for analyzing cooperative behavior in multi-agent systems. In particular, hedonic coalition formation has gained considerable attention in the literature. An interesting class of hedonic games recently introduced by Aziz et al. [3] are fractional hedonic games. In these games, the utility an agent assigns to a coalition is his average valuation for the members of his coalition. Three common notions of stability in hedonic games are core stability, Nash stability, and individual stability. For each of these notions we show that stable partitions may fail to exist in fractional hedonic games. For core stable partitions this holds even when all players only have symmetric zero/one valuations ("mutual friendship"). We then leverage these counter-examples to show that deciding the existence of stable partitions (and therefore also computing stable partitions) is NP-hard for all considered stability notions. Moreover, we show that checking whether the valuation functions of a fractional hedonic game induce strict preferences over coalitions is coNP-complete.