DAFne (Disseny d'Agents Físics)
Josep Lluís de la Rosa

y componentes del Rogi Team y del equipo de supervisión
Universidad de Girona & LEA-SICA 

Agentes físicos desarrollados como robots móviles autónomos y cooperantes
1 Motivación desde el punto de vista de aplicación industrial
1 Aplicaciones a la robótica móvil y la industria del automóvil.
De todas formas la introducción de agentes a la supervisión se podría generalizar a partir de los objetos. Las nuevas tendencias de los CASSD introducen el paradigma objeto con lo que dicha extensión parece posible y es interesante al nivel de la dificultad tecnológica de dicha extensión, su viabilidad, su funcionalidad y ver el impacto en las metodologías de diseño e implementación de sistemas de supervisión.

Motivación desde el punto de vista de investigación básica

La motivación de experimentar la IA mediante robots es la esencia misma de los agentes físicos. El hecho de utilizar robots para experimentar agentes físicos no es una decisión al azar sino que de un gran valor científico por buscar una refundación de los principios físicos o biológicos de la IA [Kitano 94, pp:34-35]. Se intenta descubrir los principios de la introspección y aprendizaje introspectivo de los agentes des de su propio fundamento físico. La construcción de sistemas inteligentes y adaptativos no es una tarea trivial. Kitano predice que la aproximación al desarrollo de un sistemas inteligente y hacerlo adaptativo sería repetir los errores de la IA tradicional, la cual supuso que la solución de ejemplos simples (toy examples) se podría generalizar a aplicaciones de mayor complejidad, expectativa la cual no se ha cumplido. Entonces creemos que para conseguir sistemas adaptativos inteligentes éstos deben presentar las características siguientes: emergencia, esto es obtener inteligencia emergiendo de las interacciones más simples entre componentes, evolución, la inteligencia es fruto de la evolución y mejora continuada de los seres inteligentes, simbiosis entre diversos componentes heterogéneos, diversidad para conseguir una correcta simbiosis y favorecer la emergencia y la evolución, motivación, la cual dirige la inteligencia, como puede ser la motivación del aprendizaje, y fundamento físico, este último de particular importancia. Insistimos en la importancia del fundamento físico dado que la inteligencia está gobernada por la física debido que los sistemas inteligentes finalmente se basan en un cuerpo físico.

Los innovadores trabajos de Luk Steels [Steels 92, 97] IJCAI, etc, van orientados por esta noción de fundamento físico de la IA y por tanto experimenta en robots aunque su investigación es en procesamiento del lenguaje natural. De una forma no tan explícita pero no por ello menos interesante [Müller 96] propone los primeros métodos para el análisis y evaluación de la mejora de rendimiento obtenida con la aplicación de agentes a robots móviles. Dentro de las nuevas tendencias en coordinación y aprendizaje de sistemas multirobots [Mataric 98] insiste en la focalización en los aspectos dinámicos de los robots (sistemas).

The integration of automatic control and mobile robotics techniques as well as the growing interest for automatic and intelligent transportation systems for social and environmental reasons are the motivations for the following proposal:

Chain of collaborative mobile robots as a test bench for automatic transportation systems

Dynamic chains made of cars, trucks or driverless vehicles will be one of the key components of some future transportation systems such as for example automatic highway systems.

Implementation of these dynamic chains are based on trajectory planning and tracking, this combined with the use of virtual links to connect the components of the chain. Trajectory planning and tracking guaranty than the chain of vehicles follow the right pass with the right speed. Virtual links ensure security distance between the vehicles and manage the vehicles adjunction or removal. The automatic supervision handles unpredictable events as apparitions of obstacles or failures of components.

In order to be able to focus on methodological development and research, we will select only one application for the project. The Automation of a small train of carts use to move luggage between airplanes and gates, this with a few risks of injuries for users. Such systems have to follow predefined paths, handle fix or mobile obstacles and be managed accordingly with the destination of the carried luggage. The difference between the proposed paradigm with current solutions is that every cart will be mechanical independent and lightly motorized. However, the motorization of the carts will not be studied in this project. This aspect is already tackled by various research groups and it is not essential for the automation. Hence, we will reproduced at a scaled size the real environment using mobile robots evolving in a airport miniature as a test bench.

>From a automatic control point of view, model based predictive control techniques will be applied and evaluated for tracking the spatial trajectory and to follow a desired speed profile. From the point of view of the computer supervision, techniques based upon collaborative agents will be implemented to deal with planned or unpredictable events and their impact along the cart  chain.

For sure, the research group of Josep Lluis de la Rosa at the University of Girona and my research group at the Swiss Federal Institute of Technology (EPFL) will participate in this project if it get funded. Any help for other R&D groups as well as industries will be welcome (INRIA, LAAS, ?).

Dr. Denis Gillet, Maître d'E&R Phone: +41 21 693-5168
Institut d'automatique FAX: +41 21 693-2574
DGM - EPFL - Ecublens email: denis.gillet@epfl.ch
CH - 1015 Lausanne Web: iawww.epfl.ch
Switzerland Telepresence: iawww2.epfl.ch

[Kitano 94] Kitano H. and Hondler J. A., "Massively Parallel Artificial Intelligence", The AAAI Press / The MIT Press pp: 1-52, 1994
[Steels 92, 97] Steels L., "The Origins of Syntax in Visually Grounded Robotic Agents", The 15 IJCAI Proceedings, pp: 1632-1641, 1997
[Mataric 98] Mataric M., "Co-ordination and Learning in Multirobot Systems", IEEE in Intelligent Systems, March/April 1998, pp: 6-9

Novedisima motivación desde el punto de vista de investigación básica en Ecosistemas Universales de Informacion.


One important question to analyse the Universal Information Ecosystems is the inclusion of physical interactions of the infoinhabitants. The physical interactions always exist as for example, the agents that trade in electronic commerce have to manage finally the delivery of the product to the human or institution that these agents represent. For instance; to have the best roasted chicken in your table at 20:00h means that your representative agent searches for the best quality/price roasted chicken in the city and automatically delivers it in these time constraint. This actions imply constrained decisions and constrained physical deliver for both software and hardware infoinhabitants. The problem is that a big concentrated demand of roasted chickens could collapse the global ecosystem. This problem is especially relevant when software decisions require of physical actions.

Important issue 1: The question is then, how these new interactions and events from the physical ecosystem and real world impact to the software ecosystem? On the other hand the decisions and interactions from the software cosystem create new initial conditions for the physical ecosystem that behaves every time differently. How these new interactions of trillions of infoinhabitants will behave in the future. Will they be safe? That is, will they be robust to minor changes? Specifically, will these new interactions between the two ecosystems unstabilise the global ecosystem?

For instance; the example of distant demands of roasted chicken is an interesting problem, that contains the complexity of travels-man problem, so that agent-based solutions are necessary:

Important Issue 2: The consensus technique from [de la Rosa] has a new parameter, initially called necessity that has potentially several meanings. Depending on the semantics (meaning) provided to the necessity important improvements thanks to more accurate vision of the co-operative world gives. The biological inspiration of this type of consensus pretends to be the most accurate consensus technique applicable to decision of infoinhabitants.

UIE seeks projects that explore novel paradigms aimed at addressing subsets of the (interrelated) challenges relevant to this initiative, and is particularly keen on research efforts that do so in an interdisciplinary manner. While the list below attempts to outline some of these research challenges, it is only illustrative and in no way prescriptive or exhaustive:

As the population of infohabitants evolves and as new activities, services, interactions and domains of discourse emerge in the ecosystem, new concepts and new language requirements can be expected to arise. Communication mechanisms as well as models used by infohabitants to reason about themselves and others will need to be amenable to evolution.
Adaptation processes include recognizing situations in which change is necessary, determining what needs to be changed, what can be changed, how to change it, with whom, etc. It also includes researching computation models and paradigms that facilitate sharing, reuse and recombination of the knowledge and expertise of infohabitants.
Note that UIE is about the underlying models for capturing and realising objectives and needs, rather than the development of new interface technologies.