Luke, S. and Spector, L., 1996, "Evolving Teamwork and Coordination with Genetic Programming." Proceedings of First Genetic Programming Conference, pp. 150-156.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....applying genetic programming techniques, namely Automatically Defined Function Genetic Programming (ADF GP) to autonomously construct an efficient agent communication protocol in a single objective agent environment. Genetic Programming is a popular method in training multiagent behavior [3][7] and has been tested In Section 2 we first briefly describe the commumcation training method proposed in our previous research [6] In Section 3 we describe our approach for applying the above training method to multiobjective agent environments. In Section 4 we describe the multiobjective Genetic ....

S.Luke and L.Spector, "Evolving Teamwork and Coordination with Genetic Programming", Genetic Programming 96, MIT Press, 1996 741

....food, or the father when it protects its territory) Role assignment therefore takes place dynamically in terms of the needs to perform specific tasks, while it is subject to continuous switching. See the work of Reynolds in [28] for competition and co evolution, and that of Luke and Spector in [45] for a discussion on the distinction between homogeneous and heterogeneous approaches in evolving multi agent systems 6. Results 6.1 Notes on Functionality The experiments conducted for this study are intended to test whether complex behaviour can emerge out of the functionality of the ....

Luke, S., Spector, L., "Evolving Teamwork and Coordination With Genetic Programming ", Proceedings of the First Annual Conference on Genetic Programming (GP-96), MIT Press, Cambridge MA, pp 150-156,

....soccer, three offensive agents are located on a rectangular field with a ball and a defensive agent. The defensive agent moves twice as quickly as the offensive agents, and the ball, when passed, moves twice as quickly as the defensive agent. This is similar to the predatorprey problem in [LS96], where more than one agent is required to solve the problem. The objective in keepaway soccer is to minimize the number of times the ball is turned over to the defender. A turnover occurs at every discrete time step in which the defender is within one grid unit of the ball. Thus, subsidiary ....

S. Luke and L. Spector. Evolving Teamwork and Coordination with Genetic Programming. In Genetic Programming 1996: Proceedings of the First Annual Conference.J.Kozaet al, eds. p. 141-149. MIT Press, Cambridge, MA, 1996.

....calculation simultaneously. 2 Evolution of Teams Haynes et al. 4] introduced the idea of team evolution into the field of genetic programming. Since then evolution of teams has been investigated mostly in connection with cooperating agents solving multi agent control problems. Luke and Spector [8] tested teamwork of homogeneous and heterogeneous agent teams in a predator prey domain and showed that the heterogenous approach is superior. In contrast to heterogenous teams homogeneous teams are composed of completely identical agents and can be evolved with the standard GP approach. Haynes et ....

....members at the same time. For recombination the participating individuals of the two parent teams can be chosen of arbitrary or equal position. If recombination between team positions is forbidden completely, the members of a team evolve independently in isolated member demes . Luke and Spector [8] showed for a control problem that team recombination restricted in this way can outperform free recombination. Isolated or semi isolated coevolution of the team members is argued to promote specialization in behaviour. A possible alternative to a random selection might be genetic operators that ....

[Article contains additional citation context not shown here]

S. Luke and L. Spector, Evolving teamwork and coordination with genetic programming. In J.R. Koza, D.E. Goldberg, David B. Fogel, and Rick L. Riolo (eds.) Genetic Programming 1996.

....tasks, i.e. a navigation problem and an escape problem. 1 Introduction Recently intelligent agents and multi agent systems have attracted much interest in Distributed Artificial Intelligence (DAI) GP and its variants have been applied to the multi agent learning (see [Haynes et al..95] Luke et al..96] Iba96] Hara et al..99] for example) However, in the multi agent application of GP, the computational burden is often problematic. This is because the number of GP trees required for the multiagent task becomes larger with the number of agents. For instance, in the heterogeneous breeding ....

....view area. This area is shown as a circle in Fig.1. If the nearest agent is out of its scope, then the Nearest Agent terminal returns a zero vector. It is also the case with the if obstacle function. There have been di#erent breeding strategies proposed for the multi agent learning by GP (see [Luke et al..96] Iba96] and [Hara et al..99] for details) This paper uses the co evolutionary breeding strategy, in which GP individuals are divided into a set of agent type subpopulations (see Fig.2) Breeding is performed in the same way as in a distributed GP. As generations proceed, some individuals are ....

Luke,S. and Spector,L., Evolving Teamwork and Coordination with Genetic Programming, in Genetic Programming 96, MIT Press, 1996

....agents. Koza and Bennett [4, 1] used genetic programming to evolve a common program that causes foraging of foods by an ant colony. Haynes et al. 2] showed that programs for solving a predator prey problem can be generated by genetic programming without any deep domain knowledge. Luke et al. [6] explored various strategies for evolving teams of agents in the Serengeti world, a simple predatorprey environment. Most of these studies have attempted to evolve emergent collective behavior immediately from primitive actions. However, more realistic complex tasks require more than one emergent ....

....An advantage of this approach is that it is easy to implement the progressive learning, i.e. learning easier tasks first and then harder tasks, which is a well proven educational method in pedagogy. It should be noted that our approach is different from other heterogeneous breeding methods [6, 3] in which different subtrees represent different agents. In the fitness switching method, different subtrees represent different behaviors of a single agent which need to be coordinated. The paper is organized as follows. Section 2 describes the table transport task and the genetic programming ....

Luke, S. and Spector, L. (1996) Evolving teamwork and coordination with genetic programming. In J.R. Koza (eds.). Proc. First Genetic Programming Conf., The MIT Press, Cambridge, MA, pp. 150-156.

....agents. Koza, 1992] and [Bennett III, 1996] used genetic programming to evolve a common program that controls foraging for food by ants. Haynes et al. 1995] showed that programs for solving a predator prey problem can be generated by genetic programming without any deep domain knowledge. [Luke and Spector, 1996] explored various strategies for evolving teams of agents in the Serengeti world, a simple predator prey environment. Iba, 1997] studied three different breeding strategies (homogeneous, heterogeneous, and coevolutionary) for cooperative robot navigation. Genetic programming was also used in ....

....by dynamically changing fitness types from a pool of fitness functions. Coevolutionary fitness switching described in this chapter is an extension of fitness switching in which multiple subtrees are coevolved in a single GP run. Our approach is different from other heterogeneous breeding schemes [Luke and Spector, 1996; Iba, 1997] in which different subtrees represent different agents. In coevolutionary fitness switching, different subtrees represent different behaviors of a single agent which need to be coordinated. The basic idea behind this approach is that fitness functions are a fundamental mechanism that ....

Luke, S. and Spector, L. (1996), "Evolving teamwork and coordination with genetic programming," in Genetic Programming 1996: Proceedings of the First Annual Conference, J. R. Koza, D. E. Goldberg, D. B. Fogel, and R. L. Riolo (Eds.), pp 150--156, Stanford University, CA, USA: MIT Press.

....independent agents working together. For instance, in the problem used in the research here, keep away soccer presents a MAS problem where one agent acting alone cannot keep a ball away from another agent who is twice as fast as the agent. Other examples are found in the predator prey problem [16] where agents (predators) try and cooperate to catch a much faster prey. MAS problems such as predator prey are used to develop solutions for MAS that can be used in many other areas. MAS problems provide research in the areas of ecient cooperation, adaptation, robustness, and real time ....

....agent moves twice as fast as the o ensive agents, and the ball can move, when passed, twice the speed of the defensive agent. When passed, the ball can travel a maximum of ten units on the eld, but the agent can kick the ball fewer if desired. This is similar to the predator prey problem in [16] where more than one agent is required to solve the problem. The problem in 21 offensive agents defensive agent path of ball Figure 3.1: The triangle of players as corners and passing lanes as edges. keep away soccer is to minimize the number of times the ball is turned over to the defender. ....

Luke, S. and L. Spector. 1996. Evolving Teamwork and Coordination with Genetic Programming. In J. Koza et al, editors, Genetic Programming 1996: Proceedings of the First Annual Conference. Cambridge. MIT Press.

....soccer three o ensive agents are located on a rectangular eld with a ball and a defensive agent. The defensive agent is twice as fast moving as the o ensive agents, and the ball can move, when passed, twice the speed of the defensive agent. This is similar to the predator prey problem in [10] where more than one agent is required to solve the problem. The problem in keep away soccer is to minimize the number of times the ball is turned over to the defender. A turnover occurs every time step that the defender is within one grid unit of the ball. Thus, the objective for o ensive agents ....

Luke, S. and L. Spector. 1996. Evolving Teamwork and Coordination with Genetic Programming. In Genetic Programming 1996: Proceedings of the First Annual Conference. J. Koza et al, eds. Cambridge: MIT Press. 141-149.

....homogeneous systems there has been a standard answer to this question. A number of researchers have applied evolutionary computation techniques to the design of various types of co operative homogeneous systems, including multi agent systems, cellular automata and parallel algorithms (e.g. Luke and Spector, 1996; Iba, 1996; Mitchell et al. 1996; Botee and Bonabeau, 1998) In all cases researchers have adopted a single population approach in which a single genotype is used to generate a system comprising identical components (i.e. clones ) This standard approach, henceforth the clonal approach , ....

Luke, S. and Spector, L. (1996). Evolving teamwork and coordination with genetic programming. In Koza, Goldberg, Fogel, and Riolo, editors, Proc. Ann. Conf. Genetic Programming. MIT Press.

....(team) and a vector of program weights form one individual and undergo evolution and tness calculation simultaneously. 2 Evolution of teams In GP the evolution of teams has been investigated mostly in connection with cooperating agents solving multi agent control problems. Luke and Spector [8] tested teamwork of homogeneous and heterogeneous agent teams in a predator prey domain and showed that the heterogenous approach is superior. In contrast to heterogenous teams homogeneous teams are composed of completely identical agents and can be evolved with the standard GP approach. In [4, 5] ....

....many team members at the same time. For recombination the participating individuals of the two parent teams can be chosen of arbitrary or equal position. If recombination between di erent team positions is not allowed, team members evolve independently in isolated member demes . Luke and Spector [8] already showed that team recombination restricted in this way can outperform free recombination. Isolated or semi isolated coevolution of the team members is argued to promote specialization in behaviour. In this contribution we do not allow recombination between di erent team positions because ....

S. Luke and L. Spector (1996) Evolving teamwork and coordination with genetic programming. In J.R. Koza, D.E. Goldberg, David B. Fogel, and Rick L. Riolo (eds.) Genetic Programming 1996: Proceedings of the First Annual Conference, 150-156, MIT Press, Cambridge, MA.

....programs evolved without the use of crossover, though they also state their belief that a wider range of tests are still required to validate their findings. Most GP work today still uses both the crossover and mutation operators. 3. 6 Teamwork with Genetic Programming Some multi agent problems[LL96] can be solved only through the use of multiple agents, others can be solved by a single, or a few single agents, but can be solved faster by multiple agents. For other problems, as more agents are assigned to solve a problem, the greater the gain in speed as each agent is assigned to the task. ....

....decision making, and heterogeneous teams where agents use different algorithms. Likewise problems can also fit into the two categories. Heterogeneous problems require heterogeneous agents to solve the problem and homogeneous problems require homogeneous problem solving agents. Luke and Spector[LL96] investigate the creation of multi agent teams using GP to solve a variety of problem types using three breeding strategies(clones, free and restricted) using a predator prey environment known as Serengeti(a grid world containing a gazelle and a group of lions) Three separate coordination ....

[Article contains additional citation context not shown here]

Sean Luke and Spector Lee. Evolving teamwork and coordination with genetic programming. In J. Koza et al, editor, Proceedings of the First Annual Conference on Genetic Programming, pages 141--149. Cambridge: MIT Press, 1996.

....than the clonal. 2 Background Arti cial evolution techniques have been applied to co operative, homogeneous systems by researchers in a number of areas, such as multi agent systems, cellular autonoma and parallel algorithms. In these cases researchers have employed the clonal approach (see e.g. [9, 7, 2]) The aclonal approach does not appear to have been considered in this context, and has typically been use to investigate game theoretical models, such as [1] in which the interests of interacting individuals con ict and homogeneity is not at issue. The clonal approach has two very obvious ....

Luke, S. and Spector, L. (1996). Evolving teamwork and coordination with Genetic Programming. In Koza et. al., editors, Genetic Programming 1996; Proc. 1 st Ann. Conf.

....tasks, i.e. a navigation problem and an escape problem. 1 Introduction Recently intelligent agents and multi agent systems have attracted much interest in Distributed Artificial Intelligence (DAI) GP and its variants have been applied to the multi agent learning (see [Haynes et al..95] Luke et al..96] Iba96] Hara et al..99] for example) However, in the multi agent application of GP, the computational burden is often problematic. This is because the number of GP trees required for the multi agent task becomes larger with the number of agents. For instance, in the heterogeneous breeding ....

....them, then evaluate the second. Else evaluate the third argument. agent is out of its scope, then the Nearest Agent terminal returns a zero vector. It is also the case with the if obstacle function. There have been di#erent breeding strategies proposed for the multi agent learning by GP (see [Luke et al..96] Iba96] and [Hara et al..99] for details) This paper uses the co evolutionary breeding strategy, in which GP individuals are divided into a set of agent type subpopulations (see Fig.2) Breeding is performed in the same way as in a distributed GP. As generations proceed, some individuals are ....

Luke,S. and Spector,L., Evolving Teamwork and Coordination with Genetic Programming, in Genetic Programming 96, MIT Press, 1996

....tasks, i.e. a navigation problem and an escape problem. 1 Introduction Recently intelligent agents and multi agent systems have attracted much interest in Distributed Artificial Intelligence (DAI) GP and its variants have been applied to the multi agent learning (see [Haynes et al..95] Luke et al..96] Iba96] Hara et al..99] for example) However, in the multi agent application of GP, the computational burden is often problematic. This is because the number of GP trees required for the multi agent task becomes larger with the number of agents. For instance, in the heterogeneous breeding ....

....view area. This area is shown as a circle in Fig.1. If the nearest agent is out of its scope, then the Nearest Agent terminal returns a zero vector. It is also the case with the if obstacle function. There have been di#erent breeding strategies proposed for the multi agent learning by GP (see [Luke et al..96] Iba96] and [Hara et al..99] for details) This paper uses the co evolutionary breeding strategy, in which GP individuals are divided into a set of agent type subpopulations (see Fig.2) Breeding is performed in the same way as in a distributed GP. As generations proceed, some individuals are ....

Luke,S. and Spector,L., Evolving Teamwork and Coordination with Genetic Programming, in Genetic Programming 96, MIT Press, 1996

....Louis, S. 860] Lozano, Manuel, 144, 309, 967, 1207] Lu, B. 59] Lu, Yilong, 1168] Lu, Y. 881] Lu, Yuchang, 1174] Lu, Zheng, 865] Luca, A. D. de P. 257] Lucas, S. M. 673] Lucas, W. K. 828] Lucasius, Carlos B. 1419] Ludvig, J. 311] Ludwig, L. A. 453] Luke, Sean, [473, 736] Lund, Donald E. 1262] Lunn, Ken, 474] Lunney, T. 513] Luong, L. H. S. 427] Luthy, Roland, 72] Lybanon, M. 1327] Lyzenga, David R. 1262] Ma, J. T. 1077, 1215, 1268] Ma, Jianhua, 657] MacDonald, R. 488] Macdonald, Rory, Macfarlane, Donald, 1421] Mach, Mari an, ....

....[146, 968, 997, 1012, 1261, 1308] Sood, Arun, 405] Soremekun, G. 732] Soule, Terence, 733, 1141] Sousa, J. M. 734] South, M. 735] Sowa, K. 110] S.Ozawa, S.Ozawa, 885] Spackman, Kent A. 1423] Spafford, Eugene H. 152] Spears, William M. 847, 1356, 1455] Spector, Lee, [473, 736, 746, 1142] Spiessens, Piet, 1456] Spiliopoulou, Myra, 745] Spittle, M. C. 328] Spofford, J. J. 1394] Spooner, E. 886] Sprave, Joachim, 1239] Sribar, Aljosa, 737, 1143] Srinivasan, Dipti, 738, 1092] Srinivasan, D. 1247] Srinivasan, V. S. 177] Sriranganathan, S. 739] ....

[Article contains additional citation context not shown here]

Sean Luke and Lee Spector. Evolving teamwork and coordination with genetic programming. In Koza et al. [1492], page ? yconf.prog ga96aLuke.

....genetic algorithm that encodes classi er systems used to control a quadruped robot; in [21] cascade neural networks [9] are evolved for parity computation using an incremental genetic algorithm. In both investigations, however, the behavioural niches of the groups are predetermined. In [14], a genetic programming approach is introduced in which niche determination is more dynamic: behaviours are evolved for a pride of lions in a predator prey task domain. Each individual s expression in the GP population codes for each and all of the behaviours required by members of the pack. The ....

....s expression. Second, the system scales with the number of agents performing the task: for n agents, the s expression must contain n branches. 2 The Model We now introduce an augmented genetic programming system, called the Legion system, which shares the advantages of the system described in [14], but overcomes its limitations. 2.1 The Legion System Each individual s expression in the Legion population encodes behaviours for an entire agent group, and is composed of two or more branch s expressions. The rst branch s expression is the partition s expression, and dictates how an agent ....

[Article contains additional citation context not shown here]

Luke, S. & L. Spector. Evolving Teamwork and Coordination with Genetic Programming. In Koza, J. R., D. E. Goldberg, D. B. Fogel & R. L. Riolo (eds.), Genetic Programming 1996: Proceedings of the First Annual Conference. MIT Press, pp. 141-149. (1996)

No context found.

Luke, S. and Spector, L., 1996, "Evolving Teamwork and Coordination with Genetic Programming." Proceedings of First Genetic Programming Conference, pp. 150-156.

No context found.

Luke, S. and Spector, L. (1996). Evolving teamwork and coordination with genetic programming. In Genetic Programming 1996: Proceedings of the First Annual Conference, pages 150--156. MIT Press.

No context found.

Sean Luke and Lee Spector. Evolving teamwork and coordination with genetic programming. In John R. Koza, David E. Goldberg, David B. Fogel, and Rick L. Riolo, editors, Genetic Programming 1996.

No context found.

S. Luke and L. Spector, "Evolving teamwork and coordination with genetic programming," in Genetic Programming 1996.

No context found.

Sean Luke and Lee Spector. Evolving teamwork and coordination with genetic programming. In John R. Koza, David E. Goldberg, David B. Fogel, and Rick L. Riolo, editors, Genetic Programming 1996: Proceedings of the First Annual Conference, pages 150--156, Stanford University, CA, USA, 28--31 July 1996. MIT Press.

No context found.

Sean Luke and Lee Spector. Evolving teamwork and coordination with genetic programming. In Genetic Programming '96, Stanford University, 1996.

No context found.

Luke, S. and Spector, L. 1996. Evolving teamwork and coordination with genetic programming. In J.R. Koza (eds.). Proc. First Genetic Programming Conf. Cambridge, MA: The MIT Press. Pages 150-156.

No context found.

Luke, S., and L. Spector. "Evolving Teamwork and Coordination with Genetic Programming ". In: Koza, John R., Goldberg, David E., Fogel, David B., and Riolo, Rick L. (eds), Genetic Programming 1996: Proceedings of the First Annual Conference, pp. 150--156. Cambridge MA: MIT Press (1996). 3

First 50 documents

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