|
|
ECAgents: Embodied and Communicating AgentsThe VisionECAgents is a project sponsored by the Future and Emerging Technologies program of the European Community (IST-FET-1940). The aim of the project is the development of a new generation of embodied agents that are able to interact directly (i.e., without human intervention) with the physical world and to communicate between them and with other agents (including humans). This will be achieved through the development of new design principles, algorithms, and mechanisms that can extend the functionality of existing technological artefacts (mobile phone, WI-FI devices, robots and robot-like artefacts, etc.) and can lead to the development of new artefacts. The project aimed to develop concepts, tools, and models for analysing collections of both natural and artificial agents, and algorithms, definitions of dynamical systems, and performance analysis tools for designing artefacts that consist of evolving populations of interacting and communicating embodied agents.
The project also aimed to investigate basic properties of different communication systems, from simple communication systems in animals to human language and technology-supported human communication, to clarify the nature of existing communications systems and to provide ideas for designing new technologies based on collections of embodied and communicating devices. The project, which is firmly rooted in the most innovative and advanced IT-technology that will become widespread in the coming 10 years, includes partners that are already doing concrete experiments with robots, wireless devices, ubiquitous environments, and living systems including humans. However, its main focus is on the development of scientific foundations by using methods, insights, and techniques from complex systems research. An evolving communication system and its underlying adaptive ontology will be viewed as a complex adaptive system, and evolutionary theory, information theory, game theory, network theory, and dynamical systems theory will all significantly contribute to its study. There is today still a tremendous gap (with some notable counter-examples) between complex systems research and IT, but this project is determined to bridge this gap for an issue of major importance. The results of the project might trigger significant breakthroughs in many future and emergent technologies, from self-developing robots to the semantic web and ubiquitous wireless devices. More specifically the project aimed at: (1) developing a new generations of embodied agents able to evolve autonomously, self-organize, and operate reliably in a dynamic environment, (2) setting up the conditions that allow a population of embodied agents to develop a shared communication language and to share knowledge, and (3) identifying new methods and algorithms that allow to engineer systems able to self-organize and to display properties emerging from the interactions between themselves and with the external environment.
The MethodIn studying the role of communication in embodied agents, the project adopted the following methodological choices: · By collections of agents is meant a plurality of agents that by interacting together exhibit collective performances that no single agent would be able to generate by acting alone. · The agents are embodied and physically situated, that is, they are physical agents interacting not only among themselves but also with the physical environment. The agents do not only exchange messages with other agents and with humans but they also move or are carried around in physical space and they interact non-symbolically with the physical environment. · The communication system of the agents is not pre-designed and is not fixed but it emerges spontaneously from the interactions of the agents among themselves and with the external environment and is in a constant state of flux due to the changing conditions of the agents, their tasks, and their environment. · Not only the communication conventions but also the underlying ontologies are assumed to self-organise and to evolve as the population of agents, the media they use, the environment, and the topics of mutual interest keep changing. · The research strategy adopted for the project is not the traditional one of studying, analysing, and experimenting with communicating agents that exist in nature but is to construct collections of artificial agents, both simulated in a computer and actual physical artefacts, and to do experiments and test hypotheses with these collections of artificial agents. · The research strategy of the project includes not only the actual construction of both simulated and real artefacts but also the study of the more abstract and general properties of large collections of interacting and communicating agents, e.g, the role of the interaction and communication network topology, the more abstract properties of communication systems (their content in information-theoretic terms, combinatiorial versus noncombinatorial systems, grammaticalized versus nongrammaticalized systems, etc.), the role of the interaction dynamics (e.g., in terms of game-theory). · Given its research strategy based on the construction of both simulated and physical artefacts, the project will not only advance our scientific understanding of already existing embodied and communicating agents but it will also suggest new technologies that consist in collections of physical devices (robots and robot-like artefacts, wireless devices, ubiquitous computing, etc.) that interact with an external environment and communicate both among themselves and with human users. Scientific and Technological ObjectivesMore specifically the project investigated the following questions: 1. Prerequisites for communication. What are the prerequisites for developing communication? How can agents enter a mode of coordinated interaction such that communication can be bootstrapped? We need to understand the cognitive prerequisites for both simple and more sophisticated communication systems, such as active and shared attention or the capacity to process hierarchical combinatorial systems, and how they can be put in place (Breazeal, 2000). From a dynamical systems point of view, coordinated interaction requires that behaviours of individual agents get entrained, for example through mechanisms like structural coupling (Ikegami and Taiji, 1999). We need to understand better the dynamical systems properties of such coupled systems (Tschacher and Dauwalter, 2001). More generally, we need to understand how emitters and receivers of signals can autonomously coordinate their activities and what role communication can play in cooperative but potentially also competitive encounters. Another important issue is how physical media can become exploited for communication. We need to understand how populations of agents can self-organise a shared repertoire of discrete building blocks grounded in a continuous physical medium and how a combinatorial system can arise from these building blocks. This is the problem of the origins of phonemic coding (for the medium of speech sounds) and syntax. 2. Conventionalised communication.
How can explicit, conventionalised, shared communication systems self-organise? What are the benefits compared to implicit communication? We need to understand how communication conventions and the processes underlying these conventions (including ontologies) can be developed autonomously and become shared in the population (Cangelosi and Parisi, 2002). What are the behaviours that have to be present in the individual units? How individual and social/communicative behaviours co-adapt and shape one another (Nolfi, 2005)? Under what conditions and with what speeds can conventions propagate so that they become shared by the total population as opposed to becoming a patchwork of local cultures (Steels, 2001)? What are the limitations in terms of size of population or flux (in and outflow of members)? What is the impact of stochasticity due to transmission errors or feedback errors? We expect that evolutionary game theory and network theory can be major sources for developing formal models that address these questions (Hammerstein and Selten, 1994; Novak and Krakauer, 1999). A key aim is to demonstrate the interplay between self-organisation, genetically based evolution, and learning from others (cultural learning) in the origins of populations with communicative systems of various degrees of complexity (Szathmary, 2001). 3. Grammatical communication.
Communication systems need to have the ability to become more complex as the need arises. In the case of human languages, this gives rise to grammatical language as the most complex form of communication in which sentences are new combinations of a basic repertory of words (lexicon), and special cues (word order, function words, etc.) are created to instruct the listener to correctly assemble the meanings of the component words into the meaning of the overall sentence. We need to understand the mechanisms by which complexity in communication can increase (Szathmary, 2001),, which includes ways for measuring this increase and its impact on success in communication or in the information processing capacity of the total system. This work has to draw both on study of evolution of natural communication systems, such as studies of grammaticalisation (Traugott and Heine, 1991), and information theory. 4. Role of communication network.
The presence and nature of communication among its components may have an important impact on the behaviour of a system. Collections of agents with sophisticated communication systems can exhibit more complex performances but they may also encounter more complex problems. For example, communicating agents create dynamical interacting networks. This emergent topology is not always beneficiary and can give rise to unexpected side effects. We have to understand the properties and efficiency of these dynamical networks and how available means for communication can impact them. Under what conditions (taking into account physical constraints, optimisation criteria, etc.) will the group adopt a broadcasting communication system or a network of peer-to-peer interactions between agents? Societies of artificial agents or humans connected through IT-devices already reach millions of members. We need the tools from complex systems theory to come to grips with the complex phenomena that arise in such circumstances. Here we expect that the recent network theories can be the proper foundation for developing prescriptive theories ( Camacho and Solé, 2001). 5. Extending the functionality of technological artefacts.
An important current trend in IT is the development of distributed technologies. A multiplicity of different artefacts (multiple computers, mobile phones, the Web, ubiquitous computing devices, etc.) already work for individual users and this trend will intensify in the near future. But these artefacts mainly communicate only with the user, not among themselves. The project will provide ideas and techniques for developing distributed technologies in which the artefacts used by a single user communicate and coordinate among themselves to provide more useful performances and services to the user. Examples of such new technologies are not only mobile robots able to communicate among themselves and with human beings (which are indeed an important area of application which we plan to investigate within this project) but also simpler sensor-equipped devices able to communicate locally among themselves through wireless connections (as in sensor webs). These embodied and communicating devices (ECDevices) will not be able to move autonomously in the environment.
However, by being carried by humans or by vehicles and by potentially affecting the behaviour of the carrier, they can interact both with the physical environment by collecting non-symbolic sensory information and by influencing the motor behaviour of the carrier, and with the social environment consisting of other ECDevices through the exchange of both symbolic and non-symbolic information. Long Term ApplicationsTechnologies, models and concepts created in the ECAgents project can have great potential to enable novel applications. As a multitude of devices with advanced computing and communication capabilities become available for everyday users, the concept of physically embodied agents that communicate and evolve through interactions between themselves and users should be applicable in many areas. Potential applications include: · Robotic systems, where agents are embodied in physical form as autonomous robots. Potential application areas include entertainment robots, service robots (e.g. collection of robots able to locate persons in need of help in devastated areas or to monitor dangerous environmental areas), and service robots companion designed to enhance and extend the individual own ability to perform crucial tasks (e.g. the ability to coordinate in order to cooperatively solve a given problem). · Ubiquitous applications where agents can move between different forms of physical embodiment. Here, agents can exist on devices such as PDA's and mobile phones. · Peer-to-peer applications where the problem is semantic interoperability with particular reference to the possibility to extend information systems with components so that peers can develop and negotiate their own communication protocols in interaction with the data world and the world of human users. This in turn might lead to the creation of an Interlingua that agents can locally interpret and in which consensus, just like in human natural languages, might emerge as a result of distributed adaptive local rules. Major Project OutcomesThe research carried within the project significantly contributed to the establishment of a new research field which investigates the evolution of communication and language in artificial embodied and situated agents. In particular, one of the mayor outcomes of the project consisted in the ability to extend the ideas developed within few pioneering studies (Cangelosi and Parisi, 1998; Di Paolo 1997, 2000; Quinn, 2001; Steels, 2001b, Steels, 2003) into a mature research area with solid theoretical and methodological foundations.
These contributions significantly progressed the state of the art with respect to our ability to develop artificial agents able to carry out cooperative and collaborative tasks by interacting directly (i.e., without human intervention) with the physical world and by communicating between them on the basis of a self-organizing communication system. More specifically the theoretical progresses achieved in the project lead to the identification of which are the pre-requisites for observing the emergence of communication skills from initially non-communicating agents (Mirolli & Parisi, 2008; Floreano et al., 2007) and for the evolution of a communication system with the characteristics of human language (Steels, 2007). Moreover, the methodological progresses (Nolfi et al. 2005) made in the project leads to the establishment of well defined methods for evolving behavioural and communicative skills from scratch serving a given functionality and for evolving a communication system with the characteristics of human language from agents provided with the required behavioural and cognitive pre-requisites. Overall, the theoretical and methodological progresses achieved in the project have enabled the possibility to scale up complexity along several dimensions including: (i) embodiment, i.e. the ability to carry on experiments on hardware and on robotic platform provided with rich sensory-motor systems (De Greeff and Nolfi, 2008, Steels, 2008), (ii) categorization and concept formation, i.e. the issue of how the meaning of signals emerge and are grounded in agents sensory-motor experiences and behavioural skills (Nolfi, 2005b; Steels & Belpaeme, 2005; Steels, 2007c), (iii) functionality, i.e. the possibility to co-evolve behavioural and communicative skills in order to solve problem which require sophisticated cooperative and/or collaborative skills (Trianni & Dorigo, 2006; Sperati et al, 2008; Ampatzis et al, 2008; De Greeff and Nolfi, 2008), (iv) expressive power and organization complexity of the self-organized communication system (De Greeff and Nolfi, 2008; Steels, 2008). The experiments performed with artificial agents also provided evidences and insights which can contribute to better modelling the evolution of communication and language in natural organisms. The data collected through these synthetic experiments represents important evidences when we consider the paucity of empirical data on the evolution animal and human communication. This shortage of empirical data is due to the impossibility to analyse the evolutionary process in action and the difficulty to reconstruct it from indirect evidence because communication and language do not leave traces in fossil records. In particular, the obtained results contributed to shed lights on how a stable and reliable communication system can evolve despite the need to develop two interdependent abilities (i.e. good signalling and good responding capabilities which are adaptively neutral by themselves) and the problems caused by the conflict of interests between individuals (see Floreano et al, 2007; Mirolli & Parisi, 2005, Mirolli & Parisi 2008).
A second major outcome of the project consisted in the study of selected aspects of different communication systems (ranging from simple communication systems in animals to human language and technology-supported human communication) aimed at the identification of important universal properties. Some of the outcomes of these research activities have already been incorporated into artificial embodied and communicating agents. Other findings have lead to the development of more abstract numerical or simulation models. Still other findings have remained at the moment at the level of theoretical elaborations. Even the latter type of results, however, could impact research aiming at studying the evolution of communication and language in embodied and situated agents in the future. Such studies include: (i) the study of chemical communication which is not only the oldest form of communication but also the most widespread communication system (from bacteria to human) and which play a key role in the transition between solitary behaviour to social behaviour and communication (Millor et al, 2006; Detrain & Deneubourg, in press); (ii) the study language dynamics and opinion dynamics in multi-agent systems through statistical physics and complex network theory methodologies (Baronchelli et al, 2006; Castellano, et al, 2007); (iii) the study of the role of communication network topologies in multi-agent systems playing language games (Dall'Asta et al. 2006); (iv) the study of complex networks which reflects the evolution of syntactic relations in human sentences during the acquisition process (Corominas-Murtra et al., 2008; Solé et al, 2008a, 2008b); (v) the study of strategic aspects of communication in humans (Hammerstein et al, 2006); (vi) the identification of selective scenarios for the origin of language Szamado & Szathmary, 2006; Hagen & Hammerstein, in press) and the theoretical analysis of the evolution of language as one of the major evolutionary transitions (Szathmary, 2008).
A third major outcome of the project consisted in progressing our understanding of the complex system nature of embodied cognition in general terms and of communication and language in particular (Steels, 2000, Nolfi, 2006; Steels, van Trijp, and Wellens, 2007). Studying behavioural and communication in embodied agents implies dealing with complex adaptive systems which are characterized by different levels of organizations involving features extending at different time scales in which, the interaction between lower-level properties lead to higher-level properties and in which higher-level properties later affect lower-levels interactions (Nolfi, 2006). More specifically for what concern the scope of the project, although communication acts are properties of a population of interacting agents which emerge from the interactions between lower level properties, they can be conceptualized as independent entities that self-organize on the basis of processes involving their constituting elements and that constraint the lower level behavioural processes from which they result. An example of self-organizing processes occurring at the level of the communication system is the competitive process between alternative words that express the same meaning (Steels, 2003). In fact, although the permanence or the disappearance of synonymous words in the communication system ultimately depends on the characteristics of individual agents and on the effects of single agent/agent communicative interactions, the destiny of synonyms can be predicted on the basis of their relative frequency of use (i.e. on the basis of a property of the communication system).
All this requires new ways to exploit bottom-up as well as top-down processes, to ensure adaptivity at all levels, to ensure the exploitation of self-organization and emergent phenomena. In other words, this requires the development of new design-for-emergence methods, such us that developed within this project (Nolfi et al., 2005). These methods are radically different from traditional design techniques based on a specification/design/test/deployment process and on a top-down divide-and-conquer methodology which attempt to decompose the overall problem into a set of supposedly simpler and independent sub-problems thus implicitly excluding the possibility to exploit properties emerging from the interactions between lower-level properties. A fourth a final major outcome of the project consisted in the preparation of a white book entitled Evolution of Communication and Language in Embodied and Situated Agents , which will be published by Springer Verlag at the end of 2008 / beginning of 2009, and in the preparation of the associated material. The book, which we hope will become a reference point in the field, aims at: (i) establishing the theoretical, methodological foundations of the field, (ii) describing the open challenges (see also Steels et al. 2004) and the most significant achievements, (iii) illustrating the scientific and technological potentials of the field, (iv) providing criteria for assessing progresses in the field, (v) illustrating the open software and hardware tools and the educational material which has been developed within the project which can allow interested readers to gain a theoretical and practical knowledge on the research conducted in the project. The book includes an introduction section that describe: (i) what embodied and communicating agents are, (ii) which are the potential impacts of this field of research for studying the evolution of language and communication in biological systems and for the development of artificial systems able to communicate and cooperate on the basis of a self-organized language, and (iii) which are the theoretical aspects beside the evolution of communication and language in embodied agents. The core of the book is constituted by section II and III which address the evolution of animal-like and human-like communication skills, respectively. Each of this section provide a description of the open challenges with respect to the current state of the art, a description of the methods which can be used to develop embodied and communicating agents, a detailed description of series of exemplificative case studies, a description of criteria for assessing research progresses, and an evaluation of the significance of the obtained results. The two concluding sections of the book illustrate an example of application of ECAgents research to an entertainment application and a brief description of the associate open-source hardware and software tools and educational material. Consortium & contact detailsThe project is coordinated by Stefano Nolfi (Institute for Cognitive Sciences and Technologies, Italian National Research Council) and includes the following partners and node coordinators:
1. National Research Council, Institute of Cognitive Sciences and Technologies (CNR-ISTC), Italy, Stefano Nolfi and Domenico Parisi. 2. Sony Computer Science Laboratory (SONY-CSL), France, Luc Steels 3. The Swiss Federal Institute of Technology Lausanne (EPFL-LIS), Switzerland, Dario Floreano 4. Université Libre de Bruxelles (ULB), Belgium, Jean-Louis Deneubourg and Marco Dorigo 5. Institute for Advanced Studies, Collegium Budapest (COLBUD), Hungary, Eörs Szathmáry 6. Future Applications Lab, University of Goteborg (VIKTORIA), Sweden, Lars Erik Holmquist 7. Humbold Universitat (UBER-ITB), Germany, Peter Hammerstein 8. Physics Department, "La Sapienza" (PHYS-SAPIENZA), Italy, Vittorio Loreto 9. University Pompeu Fabra (UPF), Spain, Ricard V. Solé 10. University of Tokyo (U-Tokyo), Japan, Takashi Ikegami Please contact: Stefano Nolfi, Institute of Cognitive Sciences and Technologies, National Research Council (CNR-ISTC) Tel: +39 06 44595233 E-mail: stefano.nolfi@istc.cnr.it ReferencesBreazeal C. (2000). Proto-conversations with an anthropomorphic robot. Proceedings of IEEE-ROMAN 2000. Workshop on Anthropomorphic Interactive Communication . Camacho J., Solé R.V. (2001). Scaling in Ecological Size Spectra. Europhysics Letters 55, 774-780 Cangelosi A, Parisi D. (Eds.) (2002). Simulating the Evolution of Language. London : Springer-Verlag. Corominas-Murtra B., Valverde S. and Solé R.V.(2008) Emergence of syntax networks through language acquisition. Journal of Theoretical Biology. Di Paolo E.A. (1997). An investigation into the evolution of communication, Adaptive Behavior 6 (2): 285-324. Di Paolo E.A. (2000). Behavioral coordination, structural congruence and entrainment in a simulation of acoustically coupled agents. Adaptive Behavior 8:(1): 25-46. Hammerstein P., Selten R. (1994). Game Theory and Evolutionary Biology, in Auman, R. and Hart, S. (eds.) Handbook of Game Theory with Economic Applications (Elsevier Science), volume 2, pp. 931-962. Hammerstein P., Hagen EH., Laubichler M. (2006). The strategic view of biological agents, Biological Theory, vol. 1, pp. 191�194 Ikegami T., Taiji M. (1999). Imitation and Cooperation in Coupled Dynamical Recognizers. In D. Floreano et al. (Eds.). Advances in Artificial Life , Springer-Verlag. Mirolli M., Parisi D. (2005): How can we explain the emergence of a language that benefits the hearer but not the speaker? Connection Science, 17(3-4): 307-324 Nolfi S., Mirolli M. (Eds). (in press). Evolution of Communication and Language in Embodied and Situated Agents. Berlin : Springer Verlag. Nowak M.A., Krakauer D. (1999). The Evolution of language , Proc. Natl. Acad. Sci. 96, 8028. Quinn M. (2001). Evolving communication without dedicated communication channels. In Kelemen, J. and Sosik, P. (Eds.) Advances in Artificial Life: Sixth European Conference on Artificial Life (ECAL 2001). Berlin: Springer Verlag. Solé R.V., Corominas-Murtra B., Valverde S. (2008b). Robust organization of evolving language networks. European Physics Journal Steels L. (2000). Language as a Complex Adaptive System. In Schoenauer, M. (Ed.), Proceedings of PPSN VI: Lecture notes in Computer Science, pages 17-26, Berlin : Springer-Verlag. Steels L. (2001) Social learning and verbal communication with humanoid robots. Proceedings of the IEEE-RAS International Conference on Humanoid Robots , Tokyo . pp. 335-342 Steels L. (2008). Modelling the formation of language. Technical report for SONY Szathmáry E. (2001). In (J. Trabant & S. Ward, Eds.): New Essays on the Origin of Language. Berlin / New York : Mouton/de Gruyter, pp. 41-51 Traugott E., Heine B. (1991). Approaches to Grammaticalization. Volume I and II . John Benjamins Publishing Company, Amsterdam . Tschacher W., Dauwalder J.P. (Eds.) (1999). Dynamics, Synergetics, Autonomous Agents. Singapore : World Scientific | ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
 |
ECAgents project started on Janury 1,2004 |
We are at month 98 of project life-time. |
web administrator: gmassera@_odiolospam_istc.cnr.it |
