Robiots: Articial and Natural Systems in Symbiosis W.W. Mayol-Cuevas (1), Jesus Savage (2), Stalin Mu~noz-Gutierrez (1), Miguel A. Villegas (2), Leoba

Transcription

1 Robiots: Articial and Natural Systems in Symbiosis W.W. Mayol-Cuevas (1), Jesus Savage (2), Stalin Mu~noz-Gutierrez (1), Miguel A. Villegas (2), Leobardo Arce (3), Gerardo Lopez (3), Horacio Ramirez (3). (1) LINDA-Group/DIE/FI UNAM (2) Departamento de Ingenieria en Computacion FI UNAM (3) Facultad de Medicina Veterinaria y Zootecnia UNAM A.P , CP Mexico DF, MEXICO wmayol@cubi210.-b.unam.mx, savage@servidor.unam.mx Tel: , FAX: Abstract Nature have been developing mechanisms that allow its species to change and evolve, one of this mechanism is symbiosis, that is when two or more species interrelate in an scheme of mutual benet. The species involved in symbiosis do not required all the structures and behavior needed for their survival, instead they share just part of them. In this paper we proposed a symbiosis between articial and natural systems that will help them in their development and performance in specic tasks. Also we proposed interaction schemes between them, and useful notation to describe these kind of systems. We will describe a current experiment showing this concept in which wearable computers for humans and no-humans are used. I. Introduction One of the problems of communication between humans and other animals is the lack of sharing a common language. We believe that an animal with the help of a computer may communicate and interact with humans using human like languages. On the other hand there have been to many approximations, theories and techniques to build articial entities with the goal to mimic nature. We believe that it is possible to obtain a better performance of these systems if we combine them with the capabilities and structures inherent in the biological beings (v.g. animals,plants), qualities that nature have give them through evolution. The scheme that we propose DO NOT correspond to the well known Cyborg one in all its current versions, however despite the fact that the objectives of the Cyborg paradigm are different from our approach we share some of its goals. We propose the introduction of symbiosis between articial and nature systems with the goal in mind that they will be both benet from this. II. Robiotic Robiotic is a term that combines the concept of Robot (or Agent) and an biological system in which both systems have a symbiotic relationship (i.e. mutual benet) [1] Robiot = (Robot + Biology) in a symbiotic relationship. Robiot is a system formed by an independent articial system and an independent biologic system. One possible example of a Robiot would be the union of a dog (with out any 1

2 surgery), in which the dog uses the computer to communicate with humans. The dog would be able to tell useful phrases for him like "Please feed me," "pet me" or "May I guide you." At the same time the computer would benet from the inherent abilities of the dog, like the dog's locomotion, pattern recognition through vision and the smell sense. In this example the computer is used as an interface of representation between the dog and the world. One way to understand the way the systems interact is by dening an operator of mutual benet: let be and two independent systems, the following notation U? > indicates that and are united but benets from. The arrow goes from the system that has the benet from the other, then the following notation <?U? > denote a mutual benet or symbiosis. Once we had dened the previous operator we may de- ne a Cyborg in the following way: Cyborg = RobotU? > Biology Cyborg = Robot <?UBiology Robiot = Robot <?U? > Biology It is important to consider the intrinsic qualities of the articial systems and nature systems to be joint in symbiosis to get the best of them. For instance the articial systems' capabilities to make fast logical like calculus as well as its easy reprogramability and data base interchanging, together with the high developed sensory systems and locomotion and navigation abilities of an animal (v.g. a dog). The above approach allows both combined systems to solve various dierent tasks in a very exible way. III. RoboCan (Robot+Computer+Animal) Currently we are developing a project, named RoboCan, that covers some of the previous ideas. The general idea is that an agent, build around a wearable computer, will direct a dog to perform basic tasks that will accomplish at the end, a specic goal. The dog will carry the wearable computer (WECO) over his body, and the computer will give the commands to the dog through a speech synthesizer, and the dog will signal the end of a task by barking, or by touching a sensor attached to one of its legs. The dog would use the WECO to express some basic feelings such as "feed me please" or "pet me", as well as the WECO will serve as a witness for life and health conditions. The main objectives of the project are the following [2]: * To develop the hardware and software necessary, using wearable computers, in order to build the following schemes: Robiot = <?U? > where and are Robiots or Cyborgs. The operator union can be extended to consider dierent degree of benet, and also to be susceptible to time and recursion. In this way a man with an articial heart is a Cyborg, and a surveillance dog with a computer that uses him to do the surveillance tour but at the same time takes him to food sources not previously known by the dog, is a Robiot. agent <?U? > dog (humanu? > CP U) <?U? > (agent <?U? > dog) * To merge the natural orientation, navigation, locomotion and well developed sensors of a dog together with the exible reprogramation

3 and data changing characteristics of a computer to obtain a Robiot. * To develop software and hardware in a WECO way, as well as new training and teaching methods for the beings involved. In respect to the above goals, we notice that our methodology in the learning stage of the dog, is substantially dierent to the popular approach, where the objective is to make that the animal learns the most as possible of instructions or commands; despite that the results could be very impressing (v.g. the Alex parrot [3]), we use a completely dierent approach: the dog will have a very reduced set of commands (nearly 20), that together with the characteristics of the computer, would have a great exibility in various dierent and may be unknown scenarios. This approach of teaching are much more less traumatic for the animal since the dog do not need to learn elaborated and much interaction commands, only a few and general commands. The basic commands given by the agent to the dog, with the following notation, (agentu? > dog) includes by the moment 3D navigational instructions, as well as commands like "take", "release", and a special instruction "free", for resting. We will activate this last instruction depending rst on the dogs' health and second on the fulllment of an instructions' queue. We are enormously interested in take great care of the dog, so the WECO will serve for stop abuse and monitoring the health of the animal and at the same time of the whole Robiot. Suppose an application of touring, RoboCan should not need to know the zone and specic route, it should suf- ce that the agent inside the WECO can modify the map database. So the system, the Robiot system is FLEXIBLE. The WECO should benet and take care of the dog as described above. One of the experiment that we are trying to resolve is the described in the following section. We think that this experiment is enough general an powerful to show the interaction of a Robiot with the world. IV. The Dispatch Machine Problem be stated as follows: Let be a special keyboard, with keys distinguished by odors, colors, etc.. (v.g. 2 dierent keys). Those keys are used to make a request of products and/or actions to the DM. The products and/or actions are obtained depending on the combination of keys; the sequence of keys is dictated by the agent to the dog, and the dog gives this sequence to the special keyboard, when the dog nish, a product and/or action is obtained. Furthermore is the possibility that the dog transports the product if any, to a specied place, following the agent's instructions. If the Robiot could solve the above DM problem, means that it may not need to know all the possible key combinations of the DM, it suces that the Robiot can push the right keys sequence to obtain those products and/or actions. More over, if all the keyboards "looks" the same, with the instruction "search," the dog can search in the place for a keyboard which could be part of very dierent machines, v.g. a T.V., a vehicle or a medical DM; and all of them could be accessed by the same Robiot. For the interaction dogu? > agent, the ergonomy should be very important, and since this area is less explored and there are nearly nothing in literature about that (i.e. wearable or computers for non-humans) this area are going to be the most experimental, and for that reason, we are now exploring various options. In answer to the question about Why a Dog, instead of a more "intelligent" animal? we can say that a dog was choose because two principal reasons: rst, because dogs are living together with mankind since various thousands of years ago, and actually the "natural" ecosystem of a dog is a city; second, because it is interesting to know what happen when the abilities of computation, processing and interaction are enhanced in an animal considered "inferior" compared to a monkey or a dolphin, whose are supposed to be "more" intelligent animals. Actually the RoboCAN project are in the stages of the construction of the designed Hardware and the dog training. The Dispatch Machine (DM) problem could

4 V. A Conjecture of Existence The Robiot paradigm, could depend of the following conjecture: Let the set of knowledge C and the receptor systems S 1 and S 2, where S 1 understands C (here understands, means that S 1 can predict, control or have the associated dynamics of a certain phenomena related to C). Then exists a translator T which allows that the combination S 2 T understands C. For the proof of the above conjecture, we think that we can use various approaches (e.g. function theory, set theory, systems theory, etc.). However we think that an existence proof is not necessary as could be a construction proof, and we are working about that. VI. Implications of the Robiotic Paradigm Some of the implications that could be the most important part of the above discussion, are the following: 6.1 For the specie which conforms the Robiot: A Robiot which would be useful to a certain society in the future, should be well appreciated inside that society. In the other hand, knowing more of a specie should always be benecial to the species involved. The benets for a certain specie inside the robiot paradigm, should be stated in dierent ways, one of them is that the computer or agent associated to a particular animal, could serve as a witness of the life conditions of the animal, could supervise its diet, stress, and in less words stop the abuse, the agents could personalize the individuals involved with, and then avoiding the dangerous uniformity of treatment to animals that in the history of humanity causes a lot of problems. More over, and with a more exiting impact, we can speculate as follows: If a certain specie is successfully involved in Robiotics, and knowing that we can change the information and/or data inside the wearable computer or agent, could we can teach or guide such specie to reproduce some of the technologies that we use to manage the world, such as seeding? or predicting some physical phenomenon? Which kind of benets will receive such specie, and which benet will receive the planet? 6.2 For the Articial Intelligence and Technology: One of the key problems inside AI in respect to the creation of articial beings, could be that such creations have not been involved inside the evolution motor or inside the natural motor. One way to do that is to place a DNA-like mechanism inside the articial creations, and to place those creations inside a colony-like structure and other biological similarities, that at least in simulation seems to give interesting results [4]; another way for making more robust articial creations is to place them inside a symbiotic relation (i.e. Robiotic). This could be viewed in practice at Nature, where symbiosis plays an important role, making that the involved beings be not so complex, because part of its surviving is shared with the others. 6.3 Between species: Suppose that there are two individuals from dierent species each one in a Robiot (R1 and R2), it seems to be possible that R1 and R2 could solve a job in a coordinated way, this job could be of benet to a human or in benet of both R1 and R2. This possibility appears exiting, since with the robiot paradigm, we then could "communicate" two species that may be not in a previous natural relation. Making that R1-2 can talk in human language, is another benet specially if R1-R2 are in constant interaction with humans (another application could be in cattle environments). The kind of training that could be useful if it is needed, should be classical or Pavlov-like and or instrumental conditioning. We are adapting the technology of software agents together with the wearable computing one that we are developing now, for two dogs and a parrot. VII. Final Remarks In this paper we present a new theoretical frame which is base of our current projects in robiotics. We present the Robiotic concept and we establish its relation to existing concepts in

5 the area. As well as we argue how the symbiotic relation between articial and biological systems could be useful and may be essential for a true realization of the articial beings in a complex world. We designed some experiments, which are base of our current projects. Some possible implications are outlined, as well as we place the important role of the wearable computing technology in the realization of the above discussion. In the conference we will present the information of the prototypes of the wearable for both humans and non-humans, together with a more detailed methodology of the RoboCAN project. References [1]W.W. Mayol, Robiotics, LINDA MEMO No May [2]J. Savage, RoboCan. Internal Report. Depto.Ing. en Computacion, FI-UNAM. January [3]I.M. Pepperberg, Conceptual abilities of some non-primate species with emphasis on an African Gray parrot. In S.T. Parker and K.R. Gibson (eds.) "language" and intelligence in monkeys and apes. pp NY, Cambridge U.P [4]K. Simms, Evolving virtual creatures, in Articial Life IV, by R. Brook and P. Maes, MIT Press, pp, pp [5]T. Starner, S. Mann, B. Rhodes, et. al. Wearable computing and augmented reality. TR No MIT Media Lab. November