Marco Dorigo, UweSchnepf, "Genetics-Based Machine Learning and BehaviourBased Robotics: A New Synthesis", IEEE Transactions on Systems, Man, and Cybernetics, January/February 1993, Vol. 23, No. 1, pp. 141-153  Home/Search   Document Details and Download   Summary   Related Articles   Check

....versions of genetic programming (GP) Koza92] to evolve programs capable of controlling the robot. Brooks92] suggests the use of GP with a high level behavioural language. Bhanzaf97] uses GP to evolve assembly code, which maps sensorial inputs into actuator actions. Rule Based Systems [Dorigo93] and [Grefenstette94] use several forms of classifier systems, or rule based systems, where the rules are genetically evolved to obtain valid controllers. Figure 1: A Genetically Programmed Network. Every node has an associated program, generated by genetic programming. In this paper we propose a ....

M. Dorigo and U. Schnepf, "Genetics- Based Machine Learning and Behaviour Based Robotics: A New Synthesis", IEEE Transactions on Systems, Man, and Cybernetics, 23, 1, 141-154, January 1993.

....of individuals using the principles of natural selection instead of more complex engineering techniques. Evolutionary approaches to agent synthesis can be divided in three main groups according with the representation used for the individuals: Neural networks [4, 8, 10] Rule based systems [6, 7, 9]. Computer programs [1, 2] The choice of the most appropriate controller architecture for autonomous agents is the center of an ongoing discussion [2, 15] which will probably never end. But from it we can conclude that the use of very specific representations has obvious disadvan tages in the ....

M. Dorigo and U. Schnepf, "Genetics-Based Machine Learning and Behaviour Based Robotics: A New Synthesis", IEEE Transactions on Systems, Man, and Cybernetics, 23, 1, 141-154, January 1993.

....[351] Comput. Ind. Eng. UK) 229] Control Engineering Practice, 90] IEE Colloq. Dig. 262] IEE Conf. Publ. ETSI konferenssi, 265] IEEE Transactions on Evolutionary Computation, 301] IEEE Transactions on Industrial Electronics, 244] IEEE Transactions on Systems, Man, and Cybernetics, [258, 264, 270, 278, 324, 325, 348] IEICE Transactions, 435] IEICE Transactions on Information and Systems, 408] Information Sciences, 311] International Journal of Vehicle Design, 290] J. Intell. Robot. Syst. Theory Appl. Netherlands) 246] J. Jpn. Soc. Precision Eng. Japan) 308] J. Robot. Syst. USA) 99, 259, ....

....[333] Czarmecki, C. 140] Czarnecki, C. 125] Daida, Jason M. 242] Dain, Robert A. 292] Davidor, Yuval, 334, 335, 336, 337, 338, 339, 340, 341, 342] Degawa, Sadao, 174] Delchambre, A. 43] Didier, K. 343] Dimou, P. 118] Doan, Chau M. 242] Dobnikar, Andrej, 133] Dorigo, Marco, [105, 111, 119, 167, 266, 278, 344, 345, 346, 347, 42, 348] Drabe, T. 33] Dubowsky, Steven, 263] Duleba, I. 104] Durantez, M. 228] Edwards, A. D. 27] Emmanuel, T. 177, 185] Enns, Russell, 168] 14 Genetic algorithms in robotics Erbudak, M. 37] Erkmen, A. M. 37] Espenschied, Kenneth S. 216] Fagg, Andrew H. 401] Falkenauer, ....

[Article contains additional citation context not shown here]

Marco Dorigo and Uwe Schnepf. Genetics-based machine learning and behaviour based robotics: A new synthesis. IEEE Transactions on Systems, Man, and Cybernetics, 23(1):141--154, 1993. ga:Dorigo93b.

.... 972] IEEE Transactions on Fuzzy Systems, 553] IEEE Transactions on Magnetics, 670] IEEE Transactions on Microwave Theory and Techniques, 717] IEEE Transactions on Power Delivery, 356] IEEE Transactions on Power Systems, 191, 849] IEEE Transactions on Systems, Man, and Cybernetics, [230, 262, 280, 471, 1041] IEICE Transactions on Fundamentals of Electronics Communications and Computer Sciences, 929] IEICE Transactions on Information and Systems, 744] Industrial and Engineering Chemistry Research, 163] Information Processing Letters, 202, 604] Information Research News, 37] Inform atica y ....

....Deodhar, D. 538] Deugo, Dwight, 220, 221] Dhawan, Atam P. 222, 223, 657, 806] Dike, B. A. 834, 837] Dissanayake, M. W. M. G. 1072] Diver, D. A. 224] Dix, T. I. 815] Dobnikar, Andrej, 225] Dockx, K. 226] Doi, Hirofumi, 1045] Dorey, Robert E. 667, 668] Dorigo, Marco, [9, 227, 228, 229, 230, 231, 232, 233] Dosi, G. 682] Duan, Q. Y. 234] Dubois, J. M. 235] Dudley, S. A. 780] Dyczij Edlinger, R. 670] East, Ian, 666] Easton, Fred F. 243] Eaton, M. 244] Edmondson, L. Vincent, 245] Ehrlich, M. 640] Eiben, Agoston E. 246, 247] Eick, C. F. 248] Eigen, Manfred, 249] ....

[Article contains additional citation context not shown here]

Marco Dorigo and Uwe Schnepf. Genetics-based machine learning and behaviour based robotics: A new synthesis. IEEE Transactions on Systems, Man, and Cybernetics, 23(1):141--154, 1993. ga:Dorigo93b.

.... Arti cial Intelligence, 85, 92] Complex Systems, 163] Computers Operations Research, 19] Design Theory and Methodology, 149] European Journal of Operational Research, 87] Evolutionary Computation, 34, 46] Expert Systems, 40] IEEE Transactions on Systems, Man, and Cybernetics, [30, 106] IEICE Transactions on Fundamentals of Electronics Communications and Computer Sciences, 175] Intelligent Systems Engineering, 137] International Journal of Intelligent Systems, 75, 143] Irish Journal of Psychology, 88] Jpn. J. Fuzzy Theory Syst. USA) 49, 71] Machine Learning, 58, ....

....41, 44, 45, 63, 65, 77] Chai, R. 172] Chalk, K. 78] Cichosz, P. 57] Collard, Philippe, 22, 48, 96] Compiani, M. 97, 98, 99, 100] Cribbs III, H. Brown, 66] Crowley, Philip H. 81] Davis, Lawrence, 101, 102] Deb, Kalyanmoy, 39, 46, 124, 125] Donnart, Jean Yves, 23] Dorigo, Marco, [38, 58, 10, 103, 104, 105, 106] Dumeur, Renaud, 107] Dwormann, Garett, 24] Endoh, Satoshi, 67] Escazut, Cathy, 22, 48] Fairley, Andrew, 25, 35, 108] Fellrath, Paul, 40] Finnerty, S. 26] Fogarty, Terence C. 20, 36, 37, 45, 47, 63, 64, 65, 77, 109] Forrest, Stephanie, 110, 111, 112, 113, 114, 115, 116, 117] Frey, ....

[Article contains additional citation context not shown here]

Marco Dorigo and Uwe Schnepf. Genetics-based machine learning and behaviour based robotics: A new synthesis. IEEE Transactions on Systems, Man, and Cybernetics, 23(1):141-154, 1993. ga:Dorigo93b.

.... for a particular environment vary considerably dependent on all the agents actions, and the best achievable profits from environment to environment also show a wide variance (making generalisation dicult) 4 AGENT STRUCTURE The agents follow a hierarchical structure similar to that described in [11]. The controller sends the current environment and the previous days rewards to two learning classi er systems, LCS1 and LCS2. Each classi er concentrates on nding rules to meet one of the objectives. LCS1 receives the reward for objective 1 (not make a loss) and LCS2 gets the reward for ....

M. Dorigo and U. Schnepf. Genetics-based machine learning and behaviour based robotics: a new synthesis. IEEE Transactions on Systems, Man and Cybernetics, SMC-23(1), 1993.

....the parallel nature of the action selection and reinforcement mechanisms. However, the cost of these advantages is in a hugely increased space of possible rules and possible overall strategies. This is one of the reasons we have adopted a complex agent structure similar to that used by Dorigo in [3], in that the agent has a high level action decision controller and learning coordination mechanism. This structure sends signals to and takes input from two classi er systems and, depending on the current agent objective and utilizing some top level long term memory structures which we call case ....

M. Dorigo and U. Schnepf. Genetics-based machine learning and behaviour based robotics: a new synthesis. IEEE Transactions on Systems, Man and Cybernetics, SMC-23(1), 1993.

....environments. Selfadaptation is then considered as a behavior based learning mechanism. All along this paper, we thus adopt a behavior based vision in our agent design. A behavior based system relies on coordination of some behavior modules, defined as primitives [Safiotti and al. 1995] [Dorigo and Schnepf 1993]. To design such a system thus consist in designing individual primitives first and then designing some kind of coordination mechanism to manage the interaction between them. However, increasing agent and task complexity can make the design difficult. Actually, in highly dynamic and in addition ....

M. Dorigo and U. Schnepf, "Genetics-based Machine Learning and Behaviour-based Robotics: A New Synthesis", in IEEE Transactions on Systems, Man and Cybernetics, 23, 1, pp. 141-154, 1993.

....These problems limit the utility of traditional model based reasoning approaches for the design of intelligent robots. Evolutionary computation provides an alternative design method that adapts the robot behavior without requiring a precisely specified model of the world [BAN95] BONA96a] BRA95][DOR93] [HOF96b] LEI96] NOL94] This paper describes our approach to evolving FLCs for a mobile robot application. We introduce a messy GA which is suitable to learn the fuzzy knowledge base according to a desired control behaviour specified by a scalar objective function. We apply our evolutionary ....

....provided by the environment the agent is able to detect which actions would result in favourable states. The agent adapts its behaviour in order to maximize the reward payoffs in the future. Evolutionary methods proved themselves as qualified for learning the behaviour of autonomous agents [BAN95][DOR93][NOL94] The previous work of Bonarini [BONA96b] Braunstingl et al. BRA95] and Leitch [LEI95] demonstrated that GAs are suitable to learn adequate fuzzy rules in order to control a mobile robot. 12 action reinforcement perception environment autonomous agent Figure 1. Interaction of the ....

M. Dorigo, U. Schnepf, "Genetics-Based Machine Learning and Behaviour-Based Robotics: A New Synthesis" IEEE Trans. on Systems, Man and Cybernetics, vol. 23, no. 1, pp. 141-154, (1993)

.... of part of the Machine Learning community who predicts that DML is today too hard a problem because Machine Learning is already so difficult in practice, pathfinders have not been stopped from studying the numerous and exciting challenges that learning in a Distributed Environment raises for AI (Dorigo and Schnepf, 1993; Grefenstette et al. 1990; Lin, 1992; Sen, 1997; Sian, 1991 ; Singh, 1992; Stone and Veloso, 1996b; Weiss, 1996 ; Weiss and Sen, 1995 ) This section supports their view and presents DML as being as inevitable in DAI as ML was for single agent AI. 3.2. General DML issues Classically, in machine ....

Dorigo, M., and Schnepf, U. 1993. Genetics-based Machine Learning and Behaviour-based Robotics : A New Synthesis. IEEE Transactions on Systems, Man, and Cybernetics, 1(23): 141-154.

....Behavior based control is based upon the concept of using simple sensorimotor processes, operating in parallel, to enable robots to react quickly and robustly in unpredictable environments. Within this general paradigm, a wide variety of different architectures have been proposed [1] 5] 7] [8] [16] 17] Like these other behavior based approaches, ELDEN uses a set of reactive behaviors to control the low level activity of the robot, but ELDEN differs in its methods for behavior activation and behavior arbitration. ELDEN s behaviors are functions that map sensory inputs to output ....

M. Dorigo and U. Schnepf, "Genetics-based machine learning and behaviour based robotics: A new synthesis," IEEE Transactions on Systems, Man, and Cybernetics, Vol. 23, No. 1, pp. 141-154, 1993.

....The challenge of embedded systems is to build controllers for large, complex, stochastic environments for which we do not possess models, and in which the agent may be hampered by incomplete perception and unreliable effectors. There are currently a mixture of methods for building such controllers [11, 18, 2, 64, 48, 25, 16, 35]. These methods vary in their degree of rigour. Early agents designed using such approaches employed controllers that were entirely reactive [2, 11] these being the only alternative 10 Open loop control is control in which the sequence of control actions is determined off line and during the ....

M. Dorigo and U. Schnepf. Genetics-based machine learning and behaviour-based robotics: A new synthesis. IEEE Transactions On Systems, Man, and Cybernetics, 23(1):141--154, January 1993.

....controller with respect to some objectives. The essential distinction of the proposed methods is the way in which the evolutionary algorithm represents the controller in the genotype. A variety of different methods, such as neural nets [20] tree structured programs [21] classifier systems [8], stimulus response rules [11] and fuzzy control rules [4] 5] 13] 18] 25] are used for the implementation of the robotic behaviour. 1.2. Soft Computing Fuzzy systems employ a mode of approximate reasoning, which allows them to make decisions based on imprecise and incomplete information in a way ....

.... called S ELF (Symbolic Evolutionary Learning of Fuzzy Rules) to learn basic robotic behaviours and to coordinate their activation [4] Dorigo et al. employed a learning classifier system to adapt behavioural patterns of a mobile robot such as light following, searching food and avoiding predators [8]. Tunstel et al. used genetic programming to learn fuzzy control rules for mobile robot path tracking[25] Braunstingl et al. optimized a fuzzy controller for a wall following behaviour of a mobile robot by means of a genetic algorithm [5] Grefenstette et al. applied their genetic learning system ....

M. Dorigo, U. Schnepf, "Genetics-Based Machine Learning and BehaviourBased Robotics: A New Synthesis", IEEE Trans. on SMC, vol. 23, no. 1, pp. 141-154, (1993).

.... without the need of a coordination structure (Steels, 1994) We believe that both these approaches are insufficient and that the process of breaking down the required behavior into sub components should be the result of an adaptation process and not of a decision of the experimenter (see also Dorigo and Schnepf, 1993). To support this hypothesis we show how in the case of a simple task which requires the ability to classify objects of different shapes, by letting the entire behavior emerge through an evolutionary technique, a more simple and robust solution can be obtained than by trying to design a set of ....

Dorigo, M., Schnepf, U. 1993. Genetic-based machine learning and behaviour based robotics: A new synthesis. IEEE transaction on Systems, Man, and Cybernetics, 23, 1, pp.141-154.

....by optimizing the robot s controller with respect to some objectives. The essential distinction of the proposed methods is the way in which the evolutionary algorithm represents the controller in the genotype. Dynamic recurrent neural nets, tree structured programs [8] classifier systems [3] or fuzzy rules [1] 6] have been employed to implement the control function. Fuzzy systems employ a mode of approximate reasoning, which allows them to make decisions based on imprecise and incomplete information in a way similar to human beings. A fuzzy system offers the advantage of knowledge ....

....design or optimization of FLCs either by learning the fuzzy if then rules or by tuning the fuzzy membership functions. Promising results have been achieved by employing evolutionary methods to develop rule based behaviors for mobile robots, given the task of prey following [1] light following [3] and collision avoidance [6] A significant part of the engineering task of designing an intelligent robot is delegated to the evolutionary algorithm, which explores alternative behaviors and optimizes the controller s parameters. 2 Mobile Robot The mobile robot is given the task of following a ....

M. Dorigo, U. Schnepf, "Genetics-Based Machine Learning and Behaviour-Based Robotics: A New Synthesis", IEEE Trans. on SMC, vol. 23, no. 1, pp. 141-154, (1993).

....to adapt. For a complex system, it should be possible to subdivide it into simpler subsystems, each with its own rule based classifier system and genetic algorithm, which would have access to the global as well as local rewards of the system. LCSs have been proposed as a model for robot control [26]. Also, LCSs have been used to construct adaptive autonomous agents [13, 101] The lawn mower problem, first considered by Koza [69] in genetic programming, is used to demonstrate an LCS in a Java applet on the Web[96] 5.6 Genetic Programming In relation to genetic algorithms, the genetic ....

M. Dorigo and U. Schnepf. Genetics-based machine learning and behaviour based robotics: A new synthesis. IEEE Trans. on Systems, Man and Cybernetics, 22(6), 1992.

....This section presents the results obtained using a trainer to help our robot, the AutonoMouse, to learn a few simple behaviors. Experiments were run using ALECSYS, a distributed version of an enhanced classifier system. Description of all the technical details regarding ALECSYS can be found in (Dorigo, 1993; 1992; Dorigo and Colombetti, 1994) and will not be discussed in this paper. Let s just say that we used a monolithic architecture (that is we used a single learning classifier system) with 300 classifiers on a three transputer configuration for the learning system, rewards and punishments ....

.... taken to control growth in behavioral complexity is behavior decomposition: the system designer identifies a set of atomic behaviors and then he uses them to build a architecture governed by some, possibly learned, control procedure (see, for example, Mahadevan and Connell, 1992; Lin, 1993; Dorigo and Schnepf, 1993; Dorigo and Colombetti, 1994) In this paper we propose a complementary approach, based on the idea of development of an agent. An agent, as it happens to animals, should go through developmental stages. In our extremely simplified model we identified three stages which differentiate in the kind ....

Dorigo, M., and Schnepf, U. (1993). Genetics-based Machine Learning and Behaviour Based Robotics: A New Synthesis. IEEE Transactions on Systems, Man, and Cybernetics, 23, 1, 141--154.

No context found.

Marco Dorigo, UweSchnepf, "Genetics-Based Machine Learning and BehaviourBased Robotics: A New Synthesis", IEEE Transactions on Systems, Man, and Cybernetics, January/February 1993, Vol. 23, No. 1, pp. 141-153

No context found.

Dorigo, M. & Schnepf, U., (1992), Genetics-based Machine Learning and Behaviour-based Robotics: A New Synthesis, IEEE Transactions on Systems Man and Cybernetics 22(6):141-154.

No context found.

Marco Dorigo, UweSchnepf, "Genetics-Based Machine Learning and BehaviourBased Robotics: A New Synthesis", IEEE Transactions on Systems, Man, and Cybernetics, January/February 1993, Vol. 23, No. 1, pp. 141-153

No context found.

Marco Dorigo and U. Schnepf. Genetics-based Machine Learning and Behaviour Based Robotics: A New Synthesis. IEEE Transactions on Systems, Man and Cybernetics, 23(1):141-154, 1993.

No context found.

Dorigo, M., and Schnepf, U. (1993). Genetics-based machine learning and behaviour based robotics: A new synthesis. IEEE Transactions on Systems, Man, and Cybernetics, 23, 141--153.

No context found.

Marco Dorigo, Uwe Schnepf, "Genetics-Based Machine Learning and BehaviourBased Robotics: A New Synthesis", IEEE Transactions on Systems, Man, and Cybernetics, January/February 1993, Vol. 23, No. 1, pp. 141-153

No context found.

Dorigo M., e U. Schnepf (1993). Genetics-based Machine Learning and Behaviour Based Robotics: A New Synthesis. IEEE Transactions on Systems, Man, and Cybernetics, 23, 1.

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