Wilson, S.W., 1985. Knowledge growth in an artificial animal, in: Grefenstette, 39 J.J., (Ed.), Proc. of ICGA 1985, Lawrence Erlabum Assoc, Hilsdale, NY, pp. 16--23.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....to store the state action list when the backward replay technique is applied. 5. Short survey of robot control tasks Studies in this area mostly deal with analyzing the environment with the aim of finding the most convenient path. One of the first studies in this area was performed by Wilson [11]. His work has led to the appearance of a whole class of tasks devoted to animats (ANIMAT = ANIMAL ROBOT) Wilson s classic animat operates and constantly learns in the same discrete environment. The task of animat is to learn to reach the goal from any starting position with the minimal number ....

Wilson S. W., "Knowledge growth in an artificial animal", //Proceedings of the First International Conference on Genetic Algorithms and their Applications, Carnegie Melon University, Pittsburgh, 1985, pp.- 1-8.

....structure of internal rules is unknown because it varies and improves in the course of learning. A system of this type learns by producing its own rule list. The rules are stored in the form of classifiers. Each classifier consists of two parts: condition part and action part. The ANIMAT problem [2] was first introduced to enable better studying of ZCS classifier systems learning abilities. A ZCS classifier system has no memory and it cannot respond to the environment messages taking into account the history of operation. It is necessary for this system to be able to identify the global ....

....the above mentioned problems. In this paper it is proposed to construct such algorithm by employing inference of finite automata using homing sequences [8] 2 ANIMAT Environment To better examine possibilities of classifier system learning, in 1985 Wilson formulated the ANIMAT learning problem [2]. The essence of this problem is searching for immovable objects in labyrinth. Agent s life consists of several cycles: it is placed in occasional place, after that the agent has to find the searched object within the least possible number of steps. The present paper makes use of this task. ....

Wilson, S. W. Knowledge growth in an artificial animal. Proceeding of the First International Conference on Genetic Algorithms and Their Applications (pp. 16-23). Hillsdale, New Jersey: Lawrence Erlbaum Associates, 1985

....as described in [8] When the bucket brigade flag is on, a payment equal to the current winner s bid (amount proportional to the classifier s strength and specificity) is given to the previous winner. This corresponds to the implicit bucket brigade algorithm first developed by S. Wilson in [23]. In general, our results show higher trader s performance when this flag is on. This is due to the fact that the payment received by the old winner balances the bid payments (and others such as taxes) it made; otherwise it could lose strength when its action might have actually contributed to ....

S. W. Wilson. Knowledge growth in an artificial animal. In John J. Grefenstette, editor, Proceedings of the 1st International Conference on Genetic Algorithms and their Applications (ICGA85), pages 16--23, Pittsburgh, PA, July 1985. Lawrence Erlbaum Associates.

....message list, and each classifier is a simple condition action pair. At each time step all classifiers matching the current input message are considered to determine an action to execute, usually through an auction. Temporal credit assignment is carried out by the implicit bucket brigade algorithm [127], a simplified version of the bucket brigade, according to which active classifiers pay their bids simply to the classifiers active at the previous time step. Such simplified classifier systems have received a significant interest. In fact, much of work with CS has been actually done with CS S ....

....version of the bucket brigade, according to which active classifiers pay their bids simply to the classifiers active at the previous time step. Such simplified classifier systems have received a significant interest. In fact, much of work with CS has been actually done with CS S GammaR , e.g. [127, 128, 121, 19, 41, 53]. Stimulus response classifier systems have potentially the same learning capabilities as the RL algorithms studied in this thesis, and therefore they deserve a special interest. In Section 6.1 the implicit bucket brigade algorithm will be closer examined and related to TD based algorithms. 2.7.3 ....

[Article contains additional citation context not shown here]

S.W. Wilson. Knowledge growth in an artificial animal. In Proceedings of International Conference on Genetic Algorithms and Their Applications, Carnegie-Mellon University, 1985.

....prediction of gR, and those i steps before the reward will converge to g i R. The speed of convergence is controlled by the learning rate parameter b (0 b 1) within the Widrow Hoff update equations. Although these processes are clearly related to traditional LCS, in particular Animat and ZCS (Wilson, 1983, 1994) the update method is novel within an LCS. In fact, the entire XCS strength formulation is novel. Each classifier carries with it not only the Prediction measure, the prediction of the average payoff it receives when invoked, but also two other related measures the Error and Accuracy ....

Wilson, S.W. (1983), Knowledge growth in an artificial animal, in Proc. First Intl. Conf. on Genetic Algorithms and their Applications, 196-201.

.... to identify a solution to the Consecutive State Problem which is less heavyweight than the more general solution proposed by Lanzi (1997, 1998) 1 INTRODUCTION XCS (Wilson, 1995, 1998) is a Learning Classifier System (Holland, 1986) with ancestry in the Animat and ZCS LCS implementations (Wilson, 1983, 1994) It maintains the basic condition action structure and ternary binary encoding of the traditional LCS with novel mechanisms for recording the strength of the classifier which separate the measure of performance utility within a given situation (payoff prediction ) from that of ....

Wilson, S.W. (1983), Knowledge growth in an artificial animal, in Proc. First Intl. Conf. on Genetic Algorithms and their Applications, 196-201.

.... [77, 78] on Machina speculatrix (an electromechanical tortoise ) almost half a century ago, biologists, psychologists and systems scientists have sought to understand links between neural hardware and behavior by building and observing simple artificial creatures animats to use Wilson s [81] term. In 1984, Braitenberg [10] published the small but now famous book Vehicles: Experiments in Synthetic Psychology. In it, he described gedanken experiments in which simple robotic creatures in spite of their simplicity displayed interesting, apparently intentional behaviors of ....

S. W. Wilson. Knowledge growth in an artificial animal. In Proceedings of 1st International Conference on Genetic Algorithms and their Applications, pages 16--23, Hillsdale, NJ, 1985. Lawrence Erlbaum.

....some of the key ideas and terminology of the new approach in Subsection 4.1, introducing New AI notions of situatedness , embodiment , and autonomous agent . Subsection 4. 2 then discusses artificial life models, focusing on one of the earliest examples of a situated artificial autonomous agent, Wilson s (1985) Animat, in order to illustrate some of the key issues in New AI in some more detail. Subsection 4.3 discusses Brooks behavior based robotics approach and his subsumption architecture. Subsection 4.4, finally, examines in detail adaptive neuro robotics, i.e. the use of artificial neural systems ....

....to question not only the techniques used by traditional AI, but also its top down approach and focus on agent internal reasoning in general. They suggested a bottom up approach, often referred to New AI or Nouvelle AI, as an alternative to the 32 framework of cognitivism and traditional AI (e.g. Wilson, 1985, 1991; Brooks, 1986a, 1990) In particular, it was agreed that AI, instead of focusing on isolated high level cognitive capacities ( micro worlds , in Dreyfus terms) should be approached first and foremost in a bottom up fashion, i.e. through the study of the interaction between simple, but ....

[Article contains additional citation context not shown here]

Wilson, Stewart W. (1985). Knowledge growth in an artificial animal. In Grefenstette, J., editor, Proceedings of an International Conference on Genetic Algorithms and Their Applications, pages 16-23. Hillsdale, NJ: Lawrence Erlbaum.

....almost cheated . Critics often say 8 when claims of emergent behaviour are made: What else could the system do But is this criticism as valid as it seems Franklin (1995, note 10, p. 207) offers what, to me, is a very satisfactory riposte to this kind of remark. Citing the objection 9 to Wilson s (1985) artificial creature Animat that: He seems . to have been jury rigged to do what his 8 I have certainly suffered this criticism in anonymous peer reviews of my own work. 9 Credited to David Lee Larom. 12 creators wanted , Franklin responds: Yes, of course. Every autonomous agent is ....

Wilson, S. W. (1985). Knowledge growth in an artificial animal. In Proceedings of 1st International Conference on Genetic Algorithms and their Applications, Hillsdale, NJ, pp. 16--23. Lawrence Erlbaum.

....Indeed, it does not use the more general bucket brigade algorithm that acts on a set of classifiers but it uses the implicit bucket brigade algorithm (Goldberg, 1989) where the currently active classifier makes a payment to the previously active classifier. This method was first developed by Wilson (1985). The currently active classifier gives a payment of b r s r to the previous active classifier. In this fashion, the expected reward in one state is backpropagated to the earlier classifiers in a behavioral chain using the ratio b r . 5 3 Some Challenges for the Anticipatory Classifier System ....

Wilson, S. W. (1985). Knowledge growth in an artificial animal. In Proceedings of an international conference on genetic algorithms and their applications (pp. 16--23). Carnegie-Mellon University, Pittsburgh, PA: John J. Grefenstette.

....Programming with pruning techniques introduced to produce human readable results (e.g. Raymer, 1997) Applying evolutionary computing to Data Mining is a more complex task, directly tackling the learning problem rather than metalearning. Early work by Parodi and Bonelli (1990) with the Animat LCS (Wilson, 1983) illustrated the potential of LCS for Data Mining within a simple data mining problem, and two significant research projects have sought to extend these results to commercial data sets (see REGAL (Giordana, 1994) and GA Miner (Flockhart, 1995) Significantly, both of these projects introduce high ....

Wilson, S.W. (1983), Knowledge Growth in an artificial animal, Proc. First Intl. Conf. On Genetic Algorithms and their Applications.

....[39] Taylor, Charles, 73, 74] Thearling, K. 23] Thompson, Adrian, 53] Thro, Ellen, 102] Todd, Peter M. 67, 98] Vaario, Jari, 24, 46, 103] Walnum, Clayton, 104] Wang, Alan, 73, 74] Watanabe, Shigeyoshi, 39] Wentworth, J. A. 38] Werner, Gregory M. 68, 69, 70] Wilson, Stewart W. [95, 96, 97, 98] Yokota, T. 28] total 92 articles by 87 different authors Subject index 13 4.7 Subject index All subject keywords of the papers given by the editor of this bibliography are shown next. The keywords neural networks , optimization , and evolution strategies have been omitted in this list ....

....of this bibliography are shown next. The keywords neural networks , optimization , and evolution strategies have been omitted in this list because of their high occurrence rate. abstract only, 105] animats, 11] review, 25] art, 17] artificial intelligence, 105, 20, 24] artificial life, [95, 96, 99, 55, 62, 97, 68, 73, 83, 89, 60, 63, 65, 71, 85, 93, 94, 100, 101, 106, 107, 56, 58, 69, 74, 77, 86, 87, 90, 91, 92, 9, 57, 59, 61, 64, 66, 67, 70, 78, 79, 80, 81, 82, 84, 88, 98, 102, 103, 104, 105, 108, 10, 11, 12, 13, 14, 15, 17, 18, 19, 20, 21, 22, 23, 25, 26, 27, 28, 29, 32, 35, 39, 40, 41, 42, 44, 47, 48, 50, 51, 52, 54] artificial life AI, 24] analysis, 43] editorial, 30] forecast, 33] modeling, 38] overview, 34] popular, 37, 49] review, 16, 46] Stanford, 75] text book, 31] artificial life , 53] autonomous systems, 52] cellular automata, 26, 27] classifier systems, 11] cognitive science, ....

[Article contains additional citation context not shown here]

Stewart W. Wilson. Knowledge growth in an artificial animal. In K. S. Narendra, editor, Adaptive and Learning Systems, page ? Plenum Press, New York, 1986. y(Wilson) ga:SWWilson86c.

....[39] Taylor, Charles, 73, 74] Thearling, K. 23] Thompson, Adrian, 53] Thro, Ellen, 102] Todd, Peter M. 67, 98] Vaario, Jari, 24, 46, 103] Walnum, Clayton, 104] Wang, Alan, 73, 74] Watanabe, Shigeyoshi, 39] Wentworth, J. A. 38] Werner, Gregory M. 68, 69, 70] Wilson, Stewart W. [95, 96, 97, 98] Yokota, T. 28] total 92 articles by 87 different authors Subject index 13 4.7 Subject index All subject keywords of the papers given by the editor of this bibliography are shown next. The keywords neural networks , optimization , and evolution strategies have been omitted in this list ....

....of this bibliography are shown next. The keywords neural networks , optimization , and evolution strategies have been omitted in this list because of their high occurrence rate. abstract only, 105] animats, 11] review, 25] art, 17] artificial intelligence, 105, 20, 24] artificial life, [95, 96, 99, 55, 62, 97, 68, 73, 83, 89, 60, 63, 65, 71, 85, 93, 94, 100, 101, 106, 107, 56, 58, 69, 74, 77, 86, 87, 90, 91, 92, 9, 57, 59, 61, 64, 66, 67, 70, 78, 79, 80, 81, 82, 84, 88, 98, 102, 103, 104, 105, 108, 10, 11, 12, 13, 14, 15, 17, 18, 19, 20, 21, 22, 23, 25, 26, 27, 28, 29, 32, 35, 39, 40, 41, 42, 44, 47, 48, 50, 51, 52, 54] artificial life AI, 24] analysis, 43] editorial, 30] forecast, 33] modeling, 38] overview, 34] popular, 37, 49] review, 16, 46] Stanford, 75] text book, 31] artificial life , 53] autonomous systems, 52] cellular automata, 26, 27] classifier systems, 11] cognitive science, ....

[Article contains additional citation context not shown here]

Stewart W. Wilson. Knowledge growth in an artificial animal. In ?, editor, Proceedings of the Fourth Yale Workshop on Applications of Adaptive Systems Theory, pages 98--104, Yale, ? 1985. Yale University, New Haven. y(Wilson) ga:SWWilson85c.

....CS has a mechanism which can perform genetic operation on a production rule (classifier as shown in Figure 1) and can acquire the knowledge desired. Also the CS can adapt to dynamic changes of the environment comparatively easily by performing genetic operations (see, for practical example Wilson[2], Hilliard et al. 3] and Goldberg[4] Thus, for CS, complete control that forecasts all behaviors of the system is not necessary. classifier : condition : action if condition then action condition =f0,1,#g L condition action =f0,1g Laction Figure 1: Definition of classifier ....

....the goal according to the score; 11. Crossover and mutate selected rule sets by genetic operations. 12. Repeat 3. 11. until at least one agent finds the path to the goal which is the certain combination of building and mark. Coding: An agent can see the grid around itself like Wilson s animat[2]. The detector translates the input message around the agent into message . Also an agent determines the direction in which to move from the output message decoded by effectors from action . Figure 10 illustrates how to encode the information to message and how to decode action to agents ....

S.W. Wilson. Knowledge growth in an artificial animal. In Proceedings of an International Conference on Genetic Algorithms and Their Applications, pages 16--23, 1985.

....artificial neural networks or evolutionary algorithms. Due to the biological inspiration and motivation underlying much of this research (cf. Sharkey and Ziemke 1998) autonomous agents are often referred to as artificial organisms , artificial life , animats (short for artificial animals ) (Wilson 1985), creatures (Brooks 1990) or biorobots (Ziemke and Sharkey 1998) These terms do not necessarily all mean exactly the same; some of them refer to physical robots only, whereas others include simple software simulations. But the terms all express the view that the mechanisms referred to are ....

....on agent internal reasoning in general. They suggested a bottom up approach, also referred to as New AI or Nouvelle AI , as an alternative to the (purely) computationalist framework of cognitivism. In particular, Brooks (1986b, 1990, 1991a) put forward his behavior based robotics approach and Wilson (1985, 1991) formulated the animat approach to AI. These approaches agree that AI should be approached from the bottom up; first and foremost through the study of the interaction between 25 autonomous agents and their environments by means of perception and action. For a more detailed review see ....

Wilson, Stewart W. (1985). Knowledge growth in an artificial animal. In Grefenstette, J., editor, Proceedings of an International Conference on Genetic Algorithms and Their Applications, pages 1623, Hillsdale, NJ. Lawrence Erlbaum Assoc.

....for solving that problem. Thus, models emerged for exploring how learning could evolve in a fixed environment [2] how language acquisition could be shaped by specific benefits of communication [3] and how simple sensory guided foraging could evolve in a world with unchanging food locations [4]. A moving target The explicit, fixed fitness function representing a fixed environment that cognition adapts to is the default assumption of original genetic algorithm models. But this assumption is more useful for doing engineering than for modeling cognitive evolution. In nature, the ....

S.W. Wilson, "Knowledge Growth in an Artificial Animal," Proc. Int'l Conf. Genetic Algorithms and their Applications, J.J. Grefenstette, ed., Lawrence Erlbaum Associates, Hillsdale, NJ 1985, pp. 16-23.

....bidding and strength adjustment could be realized. 4 Of course, many modifications of the BB and the PSP have been proposed in the literature; for instance, see the message based BB (Dorigo, 1991) the context array BB (Huang, 1989a, 1989b) the look ahead BB (Riolo, 1990) the implicit BB (Wilson, 1985), the hierarchical BB (Wilson, 1987) and the modifications of the PSP mentioned in (Grefenstette, 1988) However, none of these modifications solves the problem of lacking differentiation at the action level. orientation. A theoretical and experimental comparison of the traditional and the ....

Wilson, S.W. (1985). Knowledge growth in an artificial animal. In Proceedings of the First International Conference on Genetic Algorithms and Their Applications (pp. 16--23). Pittsburgh, PA: Erlbaum.

.... AI to a large extent overlaps with (a) the field of Adaptive Behavior (e.g. Maes, Mataric, Meyer, Pollack Wilson, 1996) which could be characterized as the study of the mechanisms of adaptive behavior in both natural and artificial agents, and (b) the Animat Approach, a term coined by Wilson (1985, 1991) since it uses artificial animals ( animats ) as tools for the study of intelligent behavior in natural ones. Furthermore these approaches, naturally, share an interest in life like or living systems with the field of Artificial Life, which is typically defined as the study of ....

Wilson, S. W. (1985). Knowledge growth in an artificial animal. Proceedings of the First International Conference on Genetic Algorithms and Their Applications (pp. 16-23). Hillsdale, NJ: Lawrence Erlbaum.

....epidemiologic data at various disease prevalence levels, and the ability of the LCS to classify previously unseen cases. The LCS used in this study follows the stimulus response model used in BOOLE (Wilson 1987) and NEWBOOLE (Bonelli et al. 1990) 2. 0 Prior studies From his work on the Animat, Wilson (1985) developed a type of stimulus response classifier system specializing in learning Boolean functions; it was accordingly named BOOLE (Wilson 1987) BOOLE has three components: performance (matching) reinforcement (credit assignment) and discovery (genetic algorithm) The performance component ....

Wilson, SW. 1985. Knowledge growth in an artificial animal. In Grefenstette, JJ (editor): Proceedings of the First International Conference on Genetic Algorithms.

....in general. They suggested a bottom up approach as an alternative to the representationalist computationalist framework of cognitivism. In particular, Brooks put forward his behaviour based AI approach [Brooks, 1986b, Brooks, 1990, Brooks, 1991a] and Wilson formulated the animat approach to AI [Wilson, 1985, Wilson, 1991] For a more detailed review see [Ziemke, 1998] In this approach, robotic agents are typically considered physically grounded as Brooks explains. Nouvelle AI is based on the physical grounding hypothesis. This hypothesis states that to build a system that is intelligent it is ....

Wilson, S. W. (1985). Knowledge growth in an artificial animal. In Grefenstette, J., editor, Proceedings of an International Conference on Genetic Algorithms and Their Applications, pages 16--23, Hillsdale, NJ. Lawrence Erlbaum Assoc.

....the animat problem. Further, computational neuroethology gives an argument for the study of artificial neural networks in closed, structured environments. 2.4. 1 The Animat Problem Simply put, the animat problem is the learning problem faced by animals and autonomous robots (collectively, animats) [46] [47] Survival (success) in an environment depends on the animal s associations between stimuli and actions leading to need satisfaction. It is presumed that some associations are genetically determined, but the remainder are learned through experiencing the environment. How can experience be ....

Stewart W. Wilson. Knowledge growth in an artificial animal. In Proceedings of an International Conference on Genetic Algorithms and their Applications, 1985.

....time and space) simulated environment, which results in the network s operation being grounded in that virtual reality. This work is a synthesis that draws from the research agendas of artificial neural networks (ANN) 4] the genetic algorithm (GA) 3] artificial life [5] the animat problem [8, 6], and computational neuroethology [2, 1] 2 MODEL AND SIMULATION A software system, AnimatE (Animat Evolution) has been developed to provide the experimenter with an environment suited for the simulation of the evolution of animats in various ecological scenarios. 2.1 Space and Time ....

Stewart W. Wilson. Knowledge growth in an artificial animal. In Grefenstette, editor, Proceedings of the First International Conference on Genetic Algorithms and their Applications, pages 16--23. Lawrence Erlbaum Associates, 1985.

....as demonstrated by the results in the next section. 5 Empirical results This section describes three environments developed by other researchers and uses the techniques from the previous section to find optimal memoryless policies for them. 5. 1 Wilson s woods7 Wilson s woods7 environment [13] consists of an 18 by 58 cell toroidal grid on which an agent can wander to any of its eight neighboring cells in a single step. Non empty cells contain trees which block motion and food which serves as the goals. An omniscient agent with complete information about its state can reach a goal ....

Stewart W. Wilson. Knowledge growth in an artificial animal. In Proceedings of the First International Conference on Genetic Algorithms and Their Applications, pages 16--23, Hillsdale, NJ, 1985. Lawrence Erlbaum Associates.

....learning architectures using genetic algorithms (GA) to dynamically adapt their knowledge representation. They handle their input messages using a population of classifiers, which are condition message rules, periodically undergoing genetic processing. The simplest kind of CS, used, e.g. by Wilson (1985), without an internal message list, is known as stimulus response classifier systems. Stimulus response classifier systems and algorithms based on the methods of temporal differences TD( Sutton, 1988) related also to dynamic programming (DP, e.g. Ross, 1983) such as AHC (Sutton, 1984) or ....

....term. For TD based algorithms this measure is typically the expected total discounted sum of reinforcement: E 1 X t=0 fl t r t # ; 1) where the discount factor 0 fl 1 adjusts the relative significance of long term rewards versus short term ones. The implicit bucket brigade algorithm (Wilson, 1985) used for credit assignment in stimulus response classifier systems may be shown to be closely related to the form of TD methods used for reinforcement learning. These relations are discussed, e.g. by Dorigo and Bersini (1994) or Cichosz (1994) This paper presents an attempt to integrate ....

Wilson, S. W. (1985). Knowledge growth in an artificial animal. In Proceedings of the International Conference on Genetic Algorithms and Their Applications.

....in a state, the maximum number of rules belonging to the sub population that matches that state is decreased. The goal is to obtain the minimum number of rules with a satisfactory performance. When the system is in a state not sufficiently covered with by any rule, a cover detector operator [25][12] generates a new rule, having the antecedent that best matches the current state, and possibly containing some don t cares . Therefore ELF may either build a rule base from scratch, or work with an initial rule base. The designer may also define constraints on the shape of the rules, and ....

S. W. Wilson, Knowledge growth in an artificial animal. Proc. of the1st Int. Conf. on Genetic Algorithms, Morgan Kaufmann, San Mateo, CA, pp. 16--23, 1985. 17

....world model, since it indicates the resulting state when taking an action in a given state. There are several known techniques for learning a goal directed world model. Among these are reinforcement learning algorithms such as genetic algorithms and the bucket brigade algorithm (Holland 1985, Wilson 1986, Booker 1988) temporal differencing techniques (Sutton Barto 1987, Sutton Barto 1989) interval estimation (Kaelbling 1990) Q learning (Watkins 1989, Sutton 1990) and variants of Q learning (Sutton 1991, Mataric 1994, Jaakkola, Jordan Singh 1994) These techniques are useful for some ....

....to demonstrate the results of this research, it seemed reasonable to use an environment that is similar to environments machine learning researchers typically use. Most research on autonomous learning is demonstrated on artificial (simulated) grid environments (see (Drescher 1989, Booker 1988, Wilson 1986, Sutton 1991) for some examples) These simulated grid environments capture some (though certainly not all) aspects of real world environments, such as office buildings or factory floors. Furthermore, they are simple to implement and to endow with any desired characteristics, such as noise or ....

Wilson, S. W. (1986), Knowledge Growth in an Artificial Animal, in K. Narendra, ed., `Adaptive and Learning Systems', Plenum Publishing Corporation.

....to produce adaptive behaviors, without human intervention, for extended periods of time. Nature is full of autonomous agents: we call them animals. However, artificial autonomous agents are much rarer things. Nevertheless they have a name: the word animat was coined for them by Stewart Wilson [42]. Animats may be real physical things, in which case they are typically autonomous mobile robots, or they may be simulated on a computer, where they go about their business in some virtual reality ; either way, there are difficult problems to be faced in constructing whole animats. The strong ....

S. W. Wilson. Knowledge growth in an artificial animal. In J. J. Grefenstette, editor, Proceedings of an International Conference on Genetic Algorithms and their Applications (ICGA85), pages 16--23, Hillsdale: New Jersey, July 1985. Lawrence Erlbaum Associates.

....1985, Forrest 1989) Simulation of Adaptive Behavior Conferences have been held in Paris, 1990 (Meyer and Wilson 1991) and Hawaii, 1992 (Meyer et al. 1993) New journals Adaptive Behavior and Artificial Life have been started in 1992 and 1993. Much work in AL has been done with simple animats (Wilson 1985), a simulated animal or autonomous robot, usually in a simple simulated environment. Evolutionary approaches are a common theme. I shall briefly summarise here just two such studies out of many possible ones of some relevance to this thesis. 2.3.1 Evolutionary Reinforcement Learning Ackley and ....

S.W. Wilson. Knowledge growth in an artificial animal. In J. Grefenstette, editor, Proceedings of the First International Conference on Genetic Algorithms and their applications, Hillsdale NJ, 1985. Lawrence Erlbaum Associates.

....medical diagnosis, theorem proving, chess playing, etc. However, these systems suffer from the problem of lacking ability to relate internal representations knowledge to the external world [7] and they tend to utterly fail when facing problems even slightly outside their domain of expertise [32] [68]. Most common drawbacks inherent to Classical AI are known as the Frame Problem (problem of maintaining a model of the real world [54] and the Grounding Problem (problem of relating the elements of the representation to the sensory information [30] Behavior based AI [41] also referred to as ....

S.W. Wilson. Knowledge growth in an artificial animal. In Proceedings of the first International conference on Genetic Algorithms and Their Applications, pages 16--23, Hillsdale, NJ, 1985. Lawrence Erlbaum Associates.

.... point of this paper is to show that this aim has to be realized, not by adopting the classical AI approach that has failed in the past, but rather by analizing and implementing a lower form of intelligence that yields broad agents [VS93] complete autonomous agents [Cli94] or animats [Wil85]. We call this approach Adaptivism. 2 Successful and Unsuccesful AI As implied by the remarks made by Gasser, quoted in section 1, DAI finds itself in a theoretical and methodological predicament. Actually this is not surprising at all, considering the state of AI itself. Attacks on classical ....

Stewart W. Wilson. Knowledge growth in an artificial animal. In J. J. Grefenstette, editor, Proceedings of the First International Conference on Genetic Algorithms and their Applications, pages 16--23, Hillsdale, NJ, 1985. Lawrence Erlbaum Associates.

....that, each time a classifier is activated its strength changes, and that this change amounts to the algebraic sum of outgoing payments (the a times strength component) environmental rewards, and ingoing payments. The VSCS reinforcement algorithm is the same as the reinforcement algorithm used in Wilson (1985) except that to create the parallel with Q learning only one classifier is reinforced on each time step. Wilson s algorithm was termed the implicit bucket brigade in Goldberg (1989) The sense of the word implicit is that there is no direct connection between a classifier activated at time t ....

Wilson S.W., 1985. Knowledge growth in an artificial animal. Proceedings of the First International Conference on Genetic Algorithms and their Applications, J.J.

....and evaluated. By adequately searching the parameter space, a good workload characterization can be found. One possible search technique is Genetic Algorithms (GAs) 9] This is a general search procedure that has had a great deal of success in many diverse problem areas (e.g. machine learning [14, 16], function optimization [2] jet engine design [6, 12] and gas pipeline control [3] This section contains a brief introduction to GAs. For a more in depth discussion see [5] A GA is a search technique that is based on the ideas found in population genetics. In order to search for an acceptable ....

S. W. Wilson, "Knowledge growth in an artificial animal, " Proceedings of an International conference on Genetic Algorithms and Their Applications, 1985, pp. 16 - 23.

.... TODO ( do what there is to do ) creature, outsmarts ANIMAT, which uses an extensive classifier system to get to know its environment; in particular when the environment is slightly changed the non learning SMARTY does better than the learning ANIMAT, no matter how long the latter s training period [18, 19]. An important, open, question is what kind of environment allows apparently purposeful behaviour to emerge due to local todo behaviour; what kind of environment leads to automatic adaptation and what type of environment makes learning a useful property 4.4. Ant and beesorting and the great ....

Wilson S. W. (1985), "Knowledge growth in an artificial animal," Genetic Algorithms, Grefenstette J. J. (ed.), Pittsburg, Pa.: Carnegie Mellon.

....ill defined, poorly constrained, and often unfriendly environments. In an effort to solve these very difficult problems facing autonomous systems, a new approach has recently emerged and is still being defined. It has been given many names by different researchers, including the animat approach [Wilson, 1985], reactive planning [Agre Chapman, 1987; Firby, 1987; Georgeff Lansky, 1987] computational neuroethology [Cliff, 1991] and the task oriented subsumption architecture [Brooks, 1986] Maes [1993] refers to these various approaches by a common underlying feature, calling it behavior based AI . ....

....change. In Proceedings of the Eleventh Annual Conference of the Cognitive Science Society, pages 940 947, Detroit, MI. Turner, R. M. 1990) A mechanism for context sensitive reasoning. Technical Report 90 68, Department of Computer Science, University of New Hampshire, Durham, NH 03824. Wilson, S. W. 1985) Knowledge growth in an artificial animal. In Proceedings of the First International Conference on Genetic ALgorithms and their Applications. Lawrence Erlbaum Associates. Zheng, X. Jackson, E. Kao, M. 1990) Object oriented sotware architecture for missionconfigurable robots. In ....

Wilson, S. W. (1985). Knowledge growth in an artificial animal. In Proceedings of the First International Conference on Genetic ALgorithms and their Applications. Lawrence Erlbaum Associates.

....tasks. The method uses multiple stimulus response type Valenzuela s FLCSs. Simple simulations for learning to steer a simulated ship are done, showing its effectiveness in fulfilling a control task and for acquiring complex control rules. Others interesting applications and studies can be found in [Wil83, Wil85, Gre87]. 6 Conclusions We have focused this paper on the description of GAs and FLCs as tools to model control processes. We have shown some GA applications to the design of FLCs, and the learning classifier systems and fuzzy learning classifier systems as tools for intelligent and adaptive control. ....

Wilson, S.W., Knowledge growth in an artificial animal. Proceedings of a International Conference on Genetic Algorithms and Their Applications, Pittsburgh, 1985, 16-23.

....place. If you want to get somewhere else, you must run at least twice as fast as that [From L. Carroll, Through The Looking Glass] It is possible to exploit this phenomenon in GA based artificial evolution to develop more and more complex competitive behaviours in animats (artificial animals [13]) without having to specify complicated evaluation functions [8] As we will see, coevolutionary GAs can also be used to tackle difficult optimisation problems of the kinds outlined in the introduction to this paper. How coevolutionary GAs should be implemented must, of course, at least in part ....

S.W. Wilson. Knowledge growth in an artificial animal. In J. Grefenstette, editor, Proceedings of the First International Conference on Genetic Algorithms and their applications. Lawrence Erlbaum Assoc., 1985.

....any goal state and null for non goal states (Barto et. al, 1989) Several laboratory scale sequential decision tasks have been investigated in the machine learning literature, including pole balancing (Selfridge, Sutton Barto, 1985) gas pipeline control (Goldberg, 1983) and the animat problem (Wilson, 1985; Wilson, 1987) In addition, sequential decision problems include many important practical problems, and much work has been devoted to their solution. The field of adaptive control theory has developed sophisticated techniques for sequential decision problems for which sufficient knowledge of the ....

....rules from a model of a particle beam accelerator, but do not explicitly address the effects of differences between the simulation model and a real target system. Goldberg (1983) describes a classifier system that learns control rules for a simulated gas pipeline. Booker (1982, 1988) and Wilson (1985) present classifier systems for organisms learning in a simulated environment. Research on classifier systems does not usually distinguish between the simulated environment used for learning and a separate target environment. One exception is Booker (1988) who discusses the possibility of an ....

[Article contains additional citation context not shown here]

Wilson, S. W. (1985). Knowledge growth in an artificial animal. Proceedings of the International Conference Genetic Algorithms and Their Applications (pp. 16-23).

....is to find a set of decision rules that maximizes the expected total payoff. 1 Several sequential decision tasks have been investigated in the machine learning literature, including pole balancing (Selfridge, Sutton Barto, 1985) gas pipeline control (Goldberg, 1983) and the animat problem (Wilson, 1985; Wilson, 1987) For many problems, including the one considered here, payoff is delayed in the sense that non null payoff occurs only at the end of an episode that may span several decision steps. In fact, the paradigm is quite broad since it includes any problem solving task by defining the ....

Wilson, S. W. (1985). Knowledge growth in an artificial animal. Proceedings of the International Conference Genetic Algorithms and Their Applications (pp. 16-23). Pittsburgh, PA.

....environments, we must define what the agent can sense and what the environment looks like. 2. 1 Environments and Agents In the literature, two types of environments have been employed to study learning classifier systems: state environments (Riolo, 1988; Smith, 1994) and woods environments (Wilson, 1985, 1994) State environments are described by directed graphs in which nodes represent the agent s sensations (so called environmental states) while edges represent the effect of actions in each state. Edges are usually labeled with the action and sometimes with a transition probability (in case ....

Wilson, S. W. (1985). Knowledge growth in an artificial animal. In Grefenstette, J. J., editor, Proceedings of the First International Conference on Genetic Algorithms and Their Applications, pages 16--23, Lawrence Erlbaum, Pittsburgh, Pennsylvania.

....the GA if the difference between that average and the current counter reading exceeds a threshold q. This technique and the deletion algorithm result in approximately equal allocation of classifiers to the various niches. Besides the GA, the discovery component contains a covering mechanism (Wilson, 1985) for 10 use in two special circumstances. First, it sometimes happens that no classifiers match a given input [M] is null. In this case, XCS simply creates a classifier with a condition matching the input and a randomly chosen action. The new classifier is inserted into [P] and a classifier ....

.... that will lead to reward, even when as with food located sparsely in the environment many actions will receive no immediate reward (food) This is the general setting of the reinforcement learning problem, and has been studied using a variety of methods, including classifier systems (e.g. Wilson, 1985), neural networks (e.g. Lin, 1993) and, especially formally, complete listings of state action pairs and their outcomes (e.g. Sutton, 1991, Watkins Dayan, 1992) In a basic kind of multi step environment, the next input y (and the reward, if any) encountered by the system depends only on the ....

[Article contains additional citation context not shown here]

Wilson, S. W. (1985). Knowledge growth in an artificial animal. Proceedings of the First International Conference on Genetic Algorithms and Their Applications (pp. 16-23). Hillsdale, New Jersey: Lawrence Erlbaum Associates.

....environments we must define what the agent can sense and what the environment looks like. 2. 1 Environments and Agents In the literature two types of environments have been employed to study learning classifier systems: state environments (Riolo 1988; Smith 1994) and woods environments (Wilson 1985; Wilson 1994) State environments are described by directed graphs in which nodes represent the agent s sensations (i.e. environmental states ) while edges represent the effect of actions in each state. Edges are usually labelled with the action and sometimes with a transition probability. The ....

Wilson, S. W. (1985). Knowledge growth in an artificial animal. In L. E. Associates (Ed.), Proceeding of the First International Conference on Genetic Algorithms and Their Applications, pp. 16--23.

....environments we must define what the agent can sense and what the environment looks like. 2. 1 Environments and Agents In the literature two types of environments have been employed to study learning classifier systems: state environments (Riolo 1988; Smith 1994) and woods environments (Wilson 1985; Wilson 1994) State environments are described by directed graphs in which nodes represent the agent s sensations (i.e. environmental states ) while edges represent the effect of actions in each state. Edges are usually labelled with the action and sometimes with a transition probability. The ....

Wilson, S. W. (1985). Knowledge growth in an artificial animal. In L. E. Associates (Ed.), Proceeding of the First International Conference on Genetic ALgorithms and Their Applications, pp. 16--23.

No context found.

Wilson, S.W., 1985. Knowledge growth in an artificial animal, in: Grefenstette, 39 J.J., (Ed.), Proc. of ICGA 1985, Lawrence Erlabum Assoc, Hilsdale, NY, pp. 16--23.

No context found.

Wilson S.W.: Knowledge Growth in an Artificial Animal, Proceedings of ICGA'85, Pittsburgh, PA, 16-23, (1985).

No context found.

S. W. Wilson. Knowledge growth in an artificial animal. In Proceedings of the First International Conference on Genetic Algorithms and their Applications, pages 16--23, Hillsdale, NJ, 1985. Lawrence Erlbaum Associates.

No context found.

S. Wilson. Knowledge growth in an artificial animal. In Proceedings of the First International Conference on Genetic Algorithms and Their Applications, pages 16--23. Hillsdale, N.J.: Erlbaum, 1985.

No context found.

Stewart W. Wilson, "Knowledge Growth in an Artificial Animal, " in Proceedings of the First International Conference on Genetic Algorithms and their Applications, edited by Greffenstette, Lawrence Erlbaum Associates, 1985.

No context found.

S.W.Wilson (1985), Knowledge Growth in an Artificial Animal, Proceedings of an International Conference on Genetic Algorithms and Their Applications, 16-23, Morgan Kaufmann.

No context found.

[Wilson, 1985] Wilson, S.W. Knowledge Growth in an artificial animal. In Proceedings of the First International Conference on Genetic Algorithms and Their Applications. 16-23. Hillsdale, NJ: Lawrence Erlbaum Associates.

No context found.

Wilson, S.W., Knowledge growth in an artificial animal. Proceedings of a International Conference on Genetic Algorithms and Their Applications, Pittsburgh, 1985, 16-23.

First 50 documents  Next 50

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC