Multi-Agent Systems and Agreement Technologies

“Perhaps a thing is simple if you can describe it fully in several different ways, without immediately knowing that you are describing the same thing” – Richard FeynmanThis articles examines multiagent learning from several paradigmatic perspectives, aiming to bring them together within one framework. We aim to provide a general definition of multiagent learning and lay out the essential characteristics of the various paradigms in a systematic manner by dissecting multiagent learning into its main components. We show how these various paradigms are related and describe similar learning processes but from varying perspectives, e.g. an individual (cognitive) learner vs. a population of (simple) learning agents.

This is a companion extended abstract to the invited talk given at EUMAS-2017. The talk was mostly based on [3, 4].

Learning to refer in a network of experts (agents) consists of distributed estimation of other experts’ topic-conditioned skills so as to refer problem instances too difficult for the referring agent to solve. This paper focuses on the cold-start case, where experts post a subset of their top skills to connected agents, and as the results show, improve overall network performance and, in particular, early-learning-phase behavior. The method surpasses state-of-the-art, i.e., proactive-DIEL, by proposing a new mechanism to penalize experts who misreport their skills, and extends the technique to other distributed learning algorithms: proactive- $$\epsilon $$ -Greedy, and proactive-Q-Learning. Our proposed new technique exhibits stronger discouragement of strategic lying, both in the limit and finite-horizon empirical analysis. The method is shown robust to noisy self-skill estimates and in evolving networks.

The evolution of the Web towards a semantically-enriched information space has risen several challenges and opportunities concerning the interaction, knowledge representation, and design of multi-agent systems. Many of these have been explored in the past, such as the usage of ontologies for defining agent knowledge bases, the definition of semantic web services, or the usage of reasoning for intelligent agent behavior. Although these efforts have resulted in important research achievements, there is still a need to provide a simple –yet comprehensive– way of interconnecting decentralized intelligent agents through a generic Web-based infrastructure. In this paper we analyze how multi-agent systems can use extensions of the Linked Data Notifications W3C recommendation as the backbone for a Semantic Web-enabled infrastructure for agent communication.

The complexity and intelligence of energy systems has increased in the recent years, whereas using Multi-agent Systems (MAS) has been recommended by IEEE for developing software solutions for modeling, controlling, and simulating their behaviors. Existing proposals on MAS solutions for energy systems proposed ad-hoc solutions for resolving specific problems, without considering interoperability and reusability. We propose a methodology, based on the Model-Driven Engineering (MDE) technique, for developing MAS solutions for energy systems. Our methodology uses the Common Information Model standard (CIM), recommended by IEEE, and the existing Platform Independent agent metamodel PIM4Agents. The proposed methodology allows modeling MAS solutions for power engineering applications, by means of a platform-independent model that abstracts developers from existing agent-oriented methodologies and platforms. Applying model transformations, the generated models can be transformed and executed within several agent platforms such as JACK and JADE. Our proposal has been validated by means of a well-known test case from the literature.

In a lot of situations a human is incapable to observe their environment properly. This can be due to disabilities, extreme conditions or simply a complex and changing environment. In those cases, help from an artificial system can be beneficial. This system, equipped with appropriate sensors, would be capable of perceiving things that a human cannot and inform them about the current state of the situation. In this short position paper, we introduce the notion of Active Situation Reporting, in which an agent can inform another agent about the evolution of a situation. We define this notion, study the challenges such a system raises and identify the open research questions by reviewing the state of the art.

We propose in this paper a distributed method to solve the security constrained optimal power flow problem (SCOPF) that considers not only contingencies on transmission lines but also on generators. With this aim, we extend the formulation of the SCOPF problem to consider the primary frequency response of generators as well as the short term constraints of generators and transmission lines. Then, we distribute the problem among different agents and we use a decentralized decision making algorithm, based on the Alternating Direction Method of Multipliers (ADMM), to optimize the grid power supply while being resilient to violations that would occur during contingencies. Finally, we validate the effectiveness of our approach on a simple test system.

Multi-Agent Systems (MAS) have been supporting the development of distributed systems performing decentralized thinking and reasoning, automated actions, and regulating component interactions in unpredictable and uncertain scenarios. Despite the scientific literature is plenty of innovative contributions about resource and tasks allocation, the agents still schedule their behaviors and tasks by employing traditional general-purpose scheduling algorithms. By doing so, MAS are unable to enforce the compliance with strict timing constraints. Thus, it is not possible to provide any guarantee about the system behavior in the worst-case scenario. Thereby, as they are, they cannot operate in safety-critical environments. This paper analyzes the agents’ local schedulers provided by the most relevant agent-based frameworks from a cyber-physical systems point of view. Moreover, it maps a set of agents’ behaviors on task models from the real-time literature. Finally, a practical case-study is provided to highlight how such “MAS reliability” can be achieved.

In this article we present theoretical results for an epistemic strategy logic with past operators, $$\text {PKSL}$$ . In $$\text {PKSL}$$ , agents are able to choose their strategies depending on past moves of other agents. This strictly extends the expressive power of some well-known epistemic strategy logics, which we illustrate by modelling forward induction: a rationality criterion, called admissibility, may be defined over agent’s strategies. Admissibility specifies coherence conditions between past and future actions, inducing new conditions for the availability of optimal strategies. We also give a resolution algorithm for $$\text {PKSL}$$ model-checking. It runs in exponential time, while the satisfiability problem is undecidable, as is the case for similar logics for strategies such as Strategy Logic.

A gossip protocol aims at arriving, by means of point-to-point communications (or telephone calls), at a situation in which every agent knows all the information initially present in the network. If it is forbidden to have more than one call at the same time, the protocol is called sequential. We generalise a method, that originates from the famous coupon collector’s problem and that was proposed by John Haigh in 1981, for bounding the expected duration of sequential gossip protocols. We give two examples of protocols where this method succeeds and two examples of protocols where this method fails to give useful bounds. Our main contribution is that, although Haigh originally applied this method in a protocol where any call is available at any moment, we show that this method can be applied in protocols where the number of available calls is decreasing. Furthermore, for one of the protocols where Haigh’s method fails we were able to obtain lower bounds for the expectation using results from random graph theory.

The paper considers term-modal logics and introduces some decidable fragments thereof. In particular, two fragments will be introduced: one that simulates monotone non-normal logics and another one that simulates normal multi-agent epistemic logics with quantification over groups of agents. These logics are defined semantically. Then, each of them is proof-theoretically characterized by a labelled calculus with good structural properties. Finally, we prove that each fragment considered is decidable, and we characterize the complexity of the validity problem for some of them.

In game theory, deciding whether a designed player wins a game corresponds to check whether he has a winning strategy. There are situations in which it is important to know whether some extra winning strategy also exists. In this paper we investigate this question over two-player turn-based games under safety and fairness objectives. We provide an automata-based technique that allows to decide in polynomial-time whether the game admits more than one winning strategy.

Nowadays economy is every day more and more a digital economy where many human activities are performed by means of digital devices. Those digital activities produce and operate on a big amount of digital assets, as the data stored in datasets, documents, images, videos or audio files. Rationally, it is useless that digital assets are made public without the specification of constrains on their usage and access. Many formal languages for expressing licenses, policies, norms, agreements, and contracts have been proposed in literature. Among them, the Open Digital Rights Language (ODRL) is a quite general one. In this paper, we present an extension of the syntax of ODRL for expressing conditional obligations. We present also an operational semantics of this extension with the goal of being able to perform automatic reasoning on the dynamic evolution in time of obligations. The definition of such operational semantics will be based on the specification of the lifecycle of obligations and on the definition of the mechanisms for computing their state using automatic reasoning. In particular, for doing that we use as far as possible, W3C standards: RDF and RDF Schema for the specification of obligations, and the Apache Jena general purpose rule engine for efficiently deducing the state of obligations on the bases of the state of the interaction among agents.

In this paper we develop a plausibility model by defining a new notion of rationality based on the assumption that a player believes that she doesn’t play a regret dominated strategy. Especially, we show that the interactive epistemic outcomes of this type of rationality are in line with the solutions of the Iterated Regret Minimization (IRM) algorithm. So, we state that one can achieve a characterization of the IRM algorithm by keeping upgrading the assumption of rationality, and we obtain common belief of rationality in the limit model. A benefit of our characterization is that it provides the epistemic foundation to the IRM algorithm and solve a dynamic information problem best expressed through the Traveler’s Dilemma. Meanwhile, we also link solutions of the IRM algorithm to modal $$\mu $$ -calculus to deepen our understanding of the epistemic characterization.

Many systems of structured argumentation explicitly require that the facts and rules that make up the argument for a conclusion be the minimal set required to derive the conclusion. $$\textsc {aspic}^{\mathsf {+}}$$ does not place such a requirement on arguments, instead requiring that every rule and fact that are part of an argument be used in its construction. Thus $$\textsc {aspic}^{\mathsf {+}}$$ arguments are minimal in the sense that removing any element of the argument would lead to a structure that is not an argument. In this paper we discuss these two types of minimality and show how the first kind of minimality can, if desired, be recovered in $$\textsc {aspic}^{\mathsf {+}}$$ .

Argumentation reasoning is a way for agents to evaluate a situation. Given a framework made of conflicting arguments, a semantics allows to evaluate the acceptability of the arguments. It may happen that the semantics associated to the framework has to be changed. In order to perform the most suitable change, the current and a potential new semantics have to be compared. Notions of difference measures between semantics have already been proposed, and application cases where they have to be minimized when a change of semantics has to be performed, have been highlighted. This paper develops these notions, it proposes an additional kind of difference measure, and shows application cases where measures may have to be maximized, and combined.

Voting rules that are based on the majority graph typically output large sets of winners. In this full original paper our goal is to investigate a general method which leads to randomized version of such rules. We use the idea of parallel universes, where each universe is connected with a permutation over alternatives. The permutation allows us to construct resolute voting rules (i.e. rules that always choose unique winners). Such resolute rules can be constructed in a variety of ways: we consider using binary voting trees to select a single alternative. In turn this permits the construction of neutral rules that output the set the possible winners of every parallel universe. The question of which rules can be constructed in this way has already been partially studied under the heading of agenda implementability. We further propose a randomised version in which the probability of being the winner is the ratio of universes in which the alternative wins. We also briefly consider (typically novel) rules that elect the alternatives that have maximal winning probability. These rules typically output small sets of winners, thus provide refinements of known tournament solutions.

Multi-agent approach has demonstrated its benefits for complex system modeling and simulation. This article focuses on how to represent and simulate a system as a set of several interacting simulators, with a focus on the case of multi-agent simulators. This raises a major challenge: multi-agent simulators are not conceived (in general) to be used with other simulators.This article presents a preliminary study about the rigorous integration of multi-agent simulators into a co-simulation platform. The work is grounded on the NetLogo simulator and the co-simulation platform mecsyco.

This paper is based on the evacuation scenario of Kiss Nightclub Tragedy in 2012. Marked by imprudence of the responsible people to the national security standards, the event has resulted in many victims. The simulations were modeled with NetLogo using Multi-Agents approach based on real data of the Nightclub and an ‘ideal’ scenario using security standard NBR 9.077 of ABNT (Brazilian National Regulamentation). The environment was modeled using Kiss blueprint. Panic was modeled using psychology basement of the literature. Results show the importance of follow the security standards imposed by ABNT to give secure evacuations of Brazilian buildings. The conclusion shows how important can be the application of this standard in the control of panic disseminate on emergency scenarios in order to provide effective evacuations.

The article addresses a lazy learning approach to fully probabilistic decision making when a decision maker (human or artificial) uses incomplete knowledge of environment and faces high computational limitations. The resulting lazy Fully Probabilistic Design (FPD) selects a decision strategy that moves a probabilistic description of the closed decision loop to a pre-specified ideal description. The lazy FPD uses currently observed data to find past closed-loop similar to the actual ideal model. The optimal decision rule of the closest model is then used in the current step. The effectiveness and capability of the proposed approach are manifested through example.

Autonomous vehicles’ behavioural analysis represents a major challenge in the automotive world. In order to ensure safety and fluidity of driving, various methods are available, in particular, simulation and formal verification. The analysis, however, has to cope with very complex environments depending on many parameters evolving in real time. In this context, none of the aforementioned approaches is fully satisfactory, which lead us to propose a combined methodology in order to point out suspicious behaviours more efficiently. We illustrate this approach by studying a non deterministic scenario involving a vehicle, which has to react to some perilous situation.

Agent based simulation of social organizations, via the investigation of agents’ training and learning tactics and strategies, has been inspired by the ability of humans to learn from social environments which are rich in agents, interactions and partial or hidden information. Such richness is a source of complexity that an effective learner has to be able to navigate. This paper focuses on the investigation of the impact of the environmental complexity on the game playing-and-learning behavior of synthetic agents. We demonstrate our approach using two independent turn-based zero-sum games as the basis of forming social events which are characterized both by competition and cooperation. The paper’s key highlight is that as the complexity of a social environment changes, an effective player has to adapt its learning and playing profile to maintain a given performance profile.

We consider pure win-lose coordination games where the representation of the game structure has additional features that are commonly known to the players, such as colouring, naming, or ordering of the available choices or of the players. We study how the information provided by such enriched representations affects the solvability of these games by means of principles of rational reasoning in coordination scenarios with no prior communication or conventions.

In classical cooperative connection situations, the agents are located at some nodes of a network and the cost of a coalition is based on the problem of finding a network of minimum cost connecting all the members of the coalition to a source.In this paper we study a different connection situation with no source and where the agents are the edges, and yet the optimal network associated to each coalition (of edges) is not fixed and follows a cost-optimization procedure. The proposed model shares some similarities with classical minimum cost spanning tree games, but also substantial differences, specifically on the appropriate way to share the costs among the agents located at the edges. We show that the core of these particular cooperative games is always non-empty and some core allocations can be easily computed.

Cooperation and negotiation are important elements of human interaction within extensive, flatly organized, mixed human-machine societies. Any sophisticated artificial intelligence cannot be complete without them. Multi-agent system with dynamic locally independent agents, that interact in a distributed way is inevitable in majority of modern applications. Here we consider a modified Ultimatum game (UG) for studying negotiation and cooperation aspects of decision making. The manuscript proposes agent’s optimizing policy using Markov decision process (MDP) framework, which covers implicit negotiation (in contrast with explicit schemes as in [5]). The proposed solution replaces the classical game-theoretical design of agents’ policies by an adaptive MDP that is: (i) more realistic with respect to the knowledge available to individual players; (ii) provides a first step towards solving negotiation essential in conflict situations.

MapReduce is a design pattern for processing large datasets on a cluster. Its performances depend on some data skews and on the runtime environment. In order to tackle these problems, we propose an adaptive multiagent system. The agents interact during the data processing and the dynamic task allocation is the outcome of negotiations. These negotiations aim at improving the workload partition among the nodes within a cluster and so decrease the runtime of the whole process. Moreover, since the negotiations are iterative the system is responsive in case of node performance variations. In this paper, we show how, when a task is divisible, an agent may split it in order to negotiate its subtasks.

Coordination of mobile multi-robot systems in a self-organised manner is in the first place beneficial for simple robots in common swarm robotics scenarios. Moreover, sophisticated robot systems as for instance in disaster rescue teams, service robotics and robot soccer can also benefit from a decentralised coordination while performing complex tasks. In order to facilitate self-organised sophisticated multi-robot applications a suitable approach is to combine individual decision-making and planning with self-organization. We introduce a framework for the implementation and application of self-organization mechanisms in multi-robot scenarios. Furthermore, the integration into the hybrid behaviour planning framework ROS Hybrid Behaviour Planner is presented. This combined approach allows for a goal-directed application of self-organisation and provides a foundation for an automated selection of suitable mechanisms.

Adaptive traffic management aims to adjust the timing of signals at road intersections to ensure smooth travel of vehicles through urban environments. A popular commercial system for handling traffic in this way is SCOOT (Split, Cycle and Offset Optimisation Technique), which involves reading data from sensors embedded in roadways to capture real-time information about traffic volume and making small changes to traffic signal timing in response. SCOOT operates in regions of connected intersections, but the sets of intersections in a region are fixed and the intersections do not communicate with each other. The research presented here takes a multi-agent approach whereby intersections work together in “coalitions” to improve traffic flow, using a market-based mechanism and forming coalitions dynamically as traffic conditions change over time. Experimental results show that this dynamic coalition approach performs better than SCOOT in several types of traffic conditions.

In this work a reverse production process is conceived as a service-based manufacturing network (ecosystem), in which the manufacturing companies “play” in the ecosystem by means of market services. One complex problem in a reverse logistic virtual market is the efficient composition and decomposition of the negotiation items. A negotiation item is defined as an item subject to be recycled: used products/scraps/wastes, a sub-part of a used product/scrap/waste, or the materials that are contained in the used product/scrap/waste. In this work we present a Multi-agent approach in order to compose the last two types of negotiation items from an orchestration of negotiation processes among the different stakeholders of the reverse logistic process (i.e. collecting points, recycling plants, disassembly plants, secondary material markets). In this way a call for buying, for example 10 tons of steel, can be handle in the virtual market as a complex process of buying and selling used products/scraps/wastes, or their sub-parts, in order to decompose and pre-process them (by recycling and/or disassembly plants) for extracting the steel contained in those items.

Intention-aware prediction is regarded as an important agreement technology to help large amount of agents in aligning their activities towards an equilibrium. If the agents do not align their activities in online routing games, then the multi-agent system is not guaranteed to get to a stable equilibrium. We formally define two intention-aware prediction methods for online routing games and empirically evaluate them in a real-world scenario. The experiments confirm that the defined intention-aware routing strategies limit the fluctuation in this online routing game scenario and make the system more or less converge to the equilibrium.

This paper proposes a crowdsourcing approach that deals with the problem of Last Mile Delivery (LMD). The proposed approach is supported by Multi Agent System (MAS) techniques and makes use of a crowd of citizens that are moving in an urban area for their own needs. The idea is to employ those citizens to deliver parcels on their way to their destinations. The complexity of the approach lies in integrating the public infrastructure network of the city for the delivery route planning, and the citizens that are deliverers in the system with their own routes to their destinations. The proposed approach is supported by a MAS framework for open fleets management. Moreover, the executed tests suggest that the LMD by citizens can drastically reduce the emissions of carbon dioxide and other airborne pollutants that are caused by delivery trucks. Moreover it can reduce the traffic congestion and noise in urban areas.

With the increase of existing sensor devices grows the data volume that is available to software systems to understand the physical world. The use of this sensor data in Multi-Agent Systems (MAS) could allow agents to improve their comprehension of the environment and provide additional information for their decision making. Unfortunately, conventional BDI agents cannot make sense of low-level sensor data directly due to their limited event comprehension capabilities: The agents react to single, isolated events rather than to multiple, related events and therefore are not able to efficiently detect complex higher-level situations from low-level sensor data. In this paper, we present Event-Driven Agents as a novel concept to enhance the perception of conventional BDI agents with Complex Event Processing. Their intended use is in environments in which percepts arrive with high speed and are too low-level to be efficiently interpreted by conventional agents directly. In a case study, we show how Event-Driven Agents can be used to address the bicycle rebalancing problem, which bike sharing systems face in their daily operations. Without an intelligent and timely intervention, bike stations of bike sharing systems tend to become empty or full quickly, which prevents the rental or return at these stations. We demonstrate how Event-Driven Agents, based on live data, can detect situations occurring in the bike sharing system in order to initiate appropriate rebalancing efforts.

Urban transportation involves a number of common problems: air and acoustic pollution, traffic jams, and so forth. This has become an important topic of study due to the interest in solving these issues in different areas (economical, social, ecological, etc.). Nowadays, one of the most popular urban transport systems are the shared vehicles systems. Among these systems there are the shared bicycle systems which have an special interest due to its characteristics. While solving some of the problems mentioned above, these systems also arise new problems such as the distribution of bicycles over time and space. Traditional approaches rely on the service provider to balancing the system, thus generating extra costs. Our proposal consists on an multi-agent system that includes user actions as a balancing mechanism, taking advantage of their trips to optimize the overall balance of the system. With this goal in mind the user is persuaded to deviate slightly from its origin/destination by providing appropriate arguments and incentives. This article presents the prediction module that will enable us to create such persuasive system. This module allow us to predict the demand for bicycles in the stations, forecasting the number of available parking spots (or available bikes). With this information the multi-agent system is capable of scoring alternative stations and routes and making offers to balance bikes across the stations. In order to achieve this, the most proper offers for the user will be predicted and used to persuade her.

An integrated network of mobile robots, personal smart devices, and smart spaces called “Robots-Assisted Ambient Intelligence” (RAmI) can provide for a more effective user assistance than if the former resources are used individually. Additionally, with the application of distributed network optimization, not only can we improve the assistance of an individual user, but we can also minimize conflict or congestion created when multiple users in large installations use the limited resources of RAmI that are spatially and temporally constrained. The emphasis of RAmI is on the efficiency and effectiveness of multiple and simultaneous user assistance and on the influence of an individual’s actions on the desired system’s performance. In this paper, we model RAmI as a multi-agent system with AmI, user, and robot agents. Moreover, we propose a modular three-layer architecture for each robot agent and discuss its application and communication requirements to facilitate efficient usage of limited RAmI resources. Our approach is showcased by means of a case study where we focus on meal and medicine delivery to patients in large hospitals.

In many real applications, to reach an agreement between the participants of a dialogue, which can be for instance a negotiation, is not easy. Indeed, there are application domains such as the medical domain where to have a consensus among medical professionals is not feasible and might even be regarded as counterproductive. In this paper, we introduce an approach for expressing goals of a dialogue considering ordered disjunction rules. By applying argumentation semantics and degrees of satisfaction of goals, we introduce the so-called dialogue agreement degree. Moreover, by considering sets of dialogue agreement degrees, we define a lattice of agreement degrees. We argue that a lattice of agreement degrees suggests different approximations between the current state of a dialogue and its aimed goals. Indeed, a lattice of agreement degrees can show evidence about whether or not it is acceptable to dismiss goals in order to maximize agreements regarding other goals.

E-Tourism applications require reliable means for sharing and reusing information and the possibility to add intelligence and inferred knowledge. In this paper, we focus on developing an ontology or common vocabulary for the tourism domain and, in particular, to represent resources from Croatia. We evaluate some of the most popular ontology development methodologies for this case. As a result of this assessment we present a proposal for a methodology that combines activities from both traditional and simplified methods.

The present work is a study of the detection of negative affective or emotional states that people have using social network sites (SNSs), and the effect that this negative state has on the repercussions of posted messages. We aim to discover in which grade an user having an affective state considered negative by an analyzer (Sentiment Analyzer and Stress Analyzer), can affect other users and generate bad repercussions, and to know whether its more suitable to predict a bad future situation using the different analyzers. We propose a method for creating a combined model of sentiment and stress and use it in our experimentation in order to discern if it is more suitable to predict future bad situations, and in what context. Additionally, we created a Multi-Agent System (MAS) that integrate the analyzers to protect or advice users, which uses the trained and tested system to predict and avoid future bad situations in social media, that could be triggered by the actions of an user that has an emotional state considered negative. We conduct this study as a way to help building future systems that prevent bad situations where an user that has a negative state creates a repercussion in the system. This can help avoid users to achieve a bad mood, or help avoid privacy issues, in the way that an user that has a negative state post information that he don’t really want to post.

Most people’s choices, including economic ones, are largely based on normative-affective considerations, not only with regard to the selection of goals but also of means. However, although emotions are inherent in human behaviour, and they are also relevant when dealing with the decision making processes, the relationship between norms and emotions has hardly been considered in the agent field, and most normative multi-agent systems do not take emotions into account, as a variable for their computation. In this paper, we analyse the advantages of including emotions in a normative system, how emotions and norms affect to each other and the work done in this field so far. To do this, we (1) identify and describe the relationships between emotions and norms; (2) review the state of art of normative emotional agents; and (3) discuss future directions for research in this field.