Agent Systems Group

Group members

• Tibor Bosse
  
• Michel Klein
  
• Peter-Paul van Maanen
• Vera Stebletsova
• Natalie van der Wal

• Charlotte Gerritsen

• Jeroen de Man
• Robbert-Jan Merk
• Gabriele Modena
• Nataliya Mogles
• Julienka Mollee
• Arlette van Wissen

• Alexei Sharpanskykh (Veni-researcher (NWO))

Specific Aspects of the Research Programme

Integrative Ambient Agent Models: Agent-Based Ambient Intelligence Applications

• Since recently Ambient Intelligence is considered a main application area for the agent systems research programme. More specifically, ambient agent models are developed that integrate domain models for human processes, and monitoring, sensor and effector capabilities. Model-based reasoning techniques are used to integrate information, in order to assess a human's functioning and to determine possible intervention actions. Applications are considered where the combination of the human's state or background and environmental demands are pressing. Examples of such circumstances are humans performing highly demanding tasks such as naval officers, fighter pilots or air traffic controllers, but also persons with difficulty in functioning in more general circumstances, for example elderly persons or persons with a chronical disease. Different cases have been addressed, varying from driver behaviour to monitoring and support of medicin intake.

Agent Modelling: Design and Simulation of Agent Systems

• Component-based agent design: In past years a method for component-based design of agent systems (DESIRE) has been developed, which includes a specification language for design specification at a conceptual and formal modelling level, and an implementation environment with a graphical interface. Methodological and semantic principles underlying the design method and the component-based agent architecture have been formulated. A large number of reusable generic component-based models has been developed and applied for different types of (software) agents in the context of Ambient Intelligence that are able to perform model-based analysis and support taking into account formalised models for human functioning, but also in other contects broker agents, negotiation agents, and process analysis and control agents.


• Hybrid causal modelling: In recent years a hybrid continuous-time-based causal modelling language LEADSTO has been developed. The language can be used for detailed specification and simulation. A dedicated software environment has been developed that supports this. As a hybrid language it integrates logical and numerical aspects, and subsumes, for example, differential equations, and executable temporal logic. It has been applied in a large number of case studies for single agent and multi-agent systems. A variety of generic models have been developed using LEADSTO, for example, for reasoning based on beliefs, desires and intentions, trust-based behaviour, intertemporal decision making and altruistic behaviour, shared extended mind, emotion and mood generation and regulation, core consciousness, criminal behaviour, reasoning by assumption, multi-representation reasoning, conditioning, eating regulation disorders, design reasoning, and bacterial behaviour.

Analysis of Agent Dynamics: Specification and Verification of Dynamic Properties

• Specification of dynamic properties: For analysis of agent systems a dedicated analysis method has been developed, in which hybrid (logical and numerical) specification of dynamics at different aggregation levels, and interlevel relations between these specifications are a central theme. Dynamic properties and interlevel relations integrate logical elements and elements from the areas of numerical analysis and calculus in Mathematics. To support this analysis method, the expressive specification language TTL (Temporal Trace Language) has been developed. This hybrid continuous time temporal predicate logical language integrates order-sorted predicate logical and numerical elements. To manage complexity the language allows for abstraction, by means of abstract language constructs that indicate complex dynamic patterns in a user-defined manner. A software environment with a graphical editor has been developed to specify dynamic properties and to check them against simulated or empirical traces. A number of templates for dynamic properties and their (interlevel) relations (proof-templates) have been identified and collected. Such examples include properties in which different possible traces are compared, which, for example, usually are not expressable within temporal logic approaches. An example of such a property is trust monotonicity: if the experiences at all past time points are more positive, the trust will be higher.


• Verification of dynamic properties: To support verification of interlevel relations between dynamic properties of different aggregation levels of the system, a software environment has been developed including SMV as model checker and transformations from the expressive TTL format into formats that can be used for model checking. Using the analysis method a number of single agent and multi-agent models have been formally modelled and analysed based on different aggregation levels and interlevel relations between them. Among the types of models analysed and verified in this manner are those for BDI-agents and adaptive agents, but also agent models for multi-attribute brokering, multi-attribute negotiation and information agents, diagnostic and design agents, assumption-based reasoning, trust-based behaviour, intertemporal decision-making, emotions, attention, criminal behaviour and organisation dynamics have been addressed.

Multidisciplinarity

Psychology

• Within the field of Cognitive Science formalisations for a number of theories and informally described models from the literature have been contributed. Among them are a dynamic model for agents behaving on the basis of beliefs, desires and intentions, and formal models and analyses for two different variants of conditioning within adaptive agents (in cooperation with cognitive psychologist S. Los). Models for assumption-based reasoning and multi-representational reasoning have been contributed, and empirically tested. A modelling approach to trust dynamics has been described and empirically tested, and a formalisation has been contributed of the philosophical interactivist perspective on mental states. In addition, a formalisation of the theory of extended mind has been contributed, models for visual attention have been developed, and a model for trust-based intertemporal decision making has been developed and analysed in its relation to the occurrence of altruist behaviour in social contexts. Finally, models for mindreading have been developed, both from the Simulation Theory and the Theory Theory perspective.


• Within the field of Clinical Psychology (in cooperation with clinical psychologist P. Cuijpers) models for emotion and mood generation and regulation have been made, and for eating regulation disorders. Moreover, single agent and multi-agent models have been developed for recurrence of depression, and the role of a person's social ties.


• Within the field of Cognitive Neuroscience, neurologist Damasio's theory of emotion, feeling and core consciousness has been formalised and a computational model for this theory has been made. Moreover, models for a human's functional state and performance have been developed. Moreover, adaptive neural models for the interaction between affective and cognitive aspects have developed, addressing, for example the interaction between beliefs or trust and induced emotions, and models for empathic understanding and the integration of cognitive and affective aspects of desires.

Social Sciences

• Within the field of Criminology (in cooperation with criminologist H. Elffers) computational formalisations of a number of well-known informal theories have been contributed, enabling computer-supported simulation experiments and formal analysis: the Routine Activity Theory, the Rational Choice Theory, the Situational Crime Prevention Theory, and the Differential Association Theory for social learning in the development of criminal behaviour.


• Within the field of Organisation Theory, contributions have been made on the extension of organisation modelling, design and analysis approaches with dynamic aspects to cover organisational behaviour in a dynamic environment at a detailed level. Furthermore, meeting protocols have been formalised and analysed. Moreover, models for different types of (centralised and decentralised) adaptive organisations have been contributed. Such models have been developed and evaluated, among others, in the context of incident management. Finally, a formal modelling approach to organisational performance indicators has been contributed.

Biology

• For the discipline Biology, in co-operation with cell biologists (J. Westerhoff, J. Snoep) first a formal model and analysis method for steady state behaviour of the cell (in particular, E. coli) in relation to its environment has been contributed (based on beliefs, desires and intentions, BDI). Moreover, the (non-steady state) dynamics of intracellular processes have been formally modelled and analysed according to a BDI-perspective and according to an organisation modelling perspective. Other biological case studies that have been addressed in a similar manner are the circulatory system in mammals and the conditioning processes of the sea hare Aplysia. Finally, the role of the development of more sophisticated cognitive systems in the evolution of behaviour to cope with increasing complexity of physical and social environments has been analysed and formalised.

Philosophy

• Within Philosophy contributions to a number of philosophical discussions have been made and formalisation has been offered. Main themes that have been addressed in this philosophical context are abstraction levels of and interlevel relations between languages in the context of reductionism, foundations of dynamics, representational content, and componential explanation. The notion of reduction relation has been further developed and formalised in order to make explicit the notion of (physical) makeup or phyisical context or embodiment of an agent. Finally, contributions have been made to the philosophical foundation of a naturalist notion of agency, in the form of criteria that determine for any process when it can be conceptualised in a faithful manner as an agent.

Foundations: Philosophical and Formal

• Philosophical foundations: Philosophical foundations for unification of diverse dynamic modelling approaches has been provided, by introducing a conceptual framework based on potentialities as anticipatory state properties and the principle of temporal factorisation, and by showing how it forms the basis for unification of, for example, cognitive modelling approaches, numerical dynamical systems modelling approaches, and logical modelling approaches.


• Formal foundations: The unified philosophical foundations of dynamics make it possible to obtain uniform conceptual semantics for hybrid modelling languages that combine numerical and logical aspects such as DESIRE, LEADSTO, and TTL and to provide a formalisation for these foundations.

Current Externally Funded Projects

• Human and cooperation modelling and support in highly demanding tasks (funded by Force Vision)


• Coordination in complex dynamic component-based software systems (funded by Force Vision)


• SOCIONICAL Project (funded under European Seventh Framework Programme (FP7))
  SOCIONICAL aims to develop Complexity Science based modelling, prediction and simulation methods for large scale socio-technical systems. SOCIONICAL focuses on the specific example of Ambient Intelligence (AmI) based smart environments. A key component of such environments is the ability to monitor user actions and to adjust its configuration and functionality accordingly. Thus, the system reacts to human behaviour while at the same influencing it. This creates a feedback loop and leads to a tight entanglement between the human and the technical system. At the same time there is dynamic, heterogeneous human-human, human-technology, and technology-technology communication leading to ad-hoc coupling between components and different feedback loops. The project will study global properties and emergent phenomena that arise in AmI based socio-technical systems from such local feedback loops and their coupling on two concrete scenarios: transportation and emergency/disaster. SOCIONICAL takes a parallel, multi facetted research approach. Thus, it will address analytical methods, complex networks based representations, and agent based models. The advances in modelling and prediction will be verified by large scale, distributed simulation driven by real life data.


• NWO ZonMW project


• SmartBandits Project (with the National Aerospace Laboratory)
  Today military missions become more and more tactical and complex. This makes integrated team training more necessary for pilots to gain the required level of tactical competencies. The expectation is that in the future the use of simulated combat situations will gain importance for training tactical competencies. In simulated combat situations the value of the contribution of computer generated forces (CGFs) is greatly affected by their ability to accurately represent human behaviour. The objective of the Smart Bandits project is to improve the tactical behaviour of the CGFs such that they can be used succesfully in simulated combat situations to train tactical competencies.


• Safety Modeling and Analysis of Organizational Processes in Air Traffic (CARE INO III Project funded by EUROCONTROL)
  In complex and distributed organisations like the air traffic industry, safe operations are the result of interactions between many entities (agents) of various types at multiple locations.In spite of the recognition of the importance of proper organizational processes for safe operations, formal modelling and simulation of organizational issues are lacking for prospective safety analysis of air traffic operations. The objective of this research project is to develop modelling and simulation methods for multi-agent organizational processes in air traffic to support proactive risk management.