Download Multi Agents System for Predator Prey Problem Progress-Computer Sciences-Project Report and more Study Guides, Projects, Research Applications of Computer Sciences in PDF only on Docsity!
i
1 Table of Contents
1 Table of Contents ..................................................................................................................... i
- 3 Introduction 2 Abstract ii
- 1.1. Multiagent System
- 1.2. Predator Prey Problem
- 3.1 Work Done
- 4 Multiagent Systems
- 4.1 Single Agent vs. Multiagent Systems
- 4.1.1 Single Agent Systems
- 4.1.2 Multiagent Systems................................................................................................
- 5 The Predator/Prey (―Pursuit‖) Domain
- 5.1 Homogeneous Non-Communicating Multiagent Systems
- 5.1.1 Homogeneous Non-Communicating Multiagent Pursuit.......................................
- 5.1.2 General Homogeneous MAS
- 5.1.3 Techniques
- 5.1.4 Issues......................................................................................................................
- 5.1.4.1 Reactive vs. Deliberative agents........................................................................
- 5.1.4.2 Local or global perspective
- 5.1.4.3 Modeling of other agents’ states........................................................................
- 5.1.4.4 How to affect others
- 5.2 Heterogeneous Non-Communicating Multiagent Systems
- 5.2.1 Heterogeneous Non-Communicating.....................................................................
- 5.2.2 General Heterogeneous MAS
- 5.2.3 Techniques
- 5.2.4 Issues....................................................................................................................
- 5.2.4.1 Benevolence vs. competitiveness
- 5.2.4.2 Stable vs. evolving agents
- 5.2.4.3 Modeling of others’ goals, actions, and knowledge
- 5.2.4.4 Resource management
- 5.2.4.5 Social conventions
- 5.2.4.6 Roles
- 5.3 Heterogeneous Communicating Multiagent Systems
- 5.3.1 Heterogeneous Communicating Multiagent Pursuit
- 5.3.2 General Communicating MAS
- 5.3.3 Techniques
- 5.3.4 Issues....................................................................................................................
- 5.3.4.1 Understanding each other
- 5.3.4.2 Planning communicative acts
- 5.3.4.3 Benevolence vs. competitiveness
- 5.3.4.4 Resource management
- 5.3.4.5 Commitment/decommitment
- 6 References
ii
2 Abstract
Multi Agents System for Predator Prey Problem is a simulator which is used for the simulations of Predator prey Problem. Simulators are used in various field because they are cheaper and produce faster results. In Multi Agent System Simulators are used for testing the strategy of completing the task. Here progress of the project up to the Mid of 8th semester is described which consist of Control Algorithm for the predator Prey Problem.
3.1 Work Done
At the end of 7th^ semester Map Representation and Path Planning were completed and presented. In this report User defined control Algorithm is explained which will be given in the GUI. After this the work left is the path planning for the Multi Agent System, Building GUI and thesis writing.
4 Multiagent Systems
Two obvious questions about any type of technology are: What advantages does it offer over the alternatives? In what circumstances is it useful? It would be foolish to claim that MAS should be used when designing all complex systems. Like any useful approach, there are some situations for which it is particularly appropriate, and others for which it is not. The goal of this section is to underscore the need for and usefulness of MAS while giving characteristics of typical domains that can benefit from it. The most important reason to use MAS when designing a system is that some domains require it. In particular, if there are different people or organizations with different (possibly conflicting) goals and proprietary information, then a multiagent system is needed to handle their interactions. Even if each organization wants to model its internal affairs with a single system, the organizations will not give authority to any single person to build a system that represents them all: the different organizations will need their own systems that reflect their capabilities and priorities. For example of a domain that requires MAS is hospital scheduling as presented in [1]. This domain from an actual case study requires different agents to represent the interests of different people within the hospital. Hospital employees have different interests, from nurses who want to minimize the patient’s time in the hospital, to x-ray operators who want to maximize the throughput on their machines. Since different people evaluate candidate schedules with different criteria, they must be represented by separate agents if their interests are to be justly considered. Even in domains that could conceivably use systems that are not distributed, there are several possible reasons to use MAS. Having multiple agents could speed up a system’s operation by providing a method for parallel computation. For instance, a domain that is easily broken into components could benefit from MAS. Furthermore, the parallelism of MAS can help deal with limitations imposed by time bounded reasoning requirements. While parallelism is achieved by assigning different tasks or abilities to different agents, robustness is a benefit of multiagent systems that have redundant agents. If control and responsibilities are sufficiently shared among different agents, the system can tolerate failures by one or more of the agents. Domains that must degrade gracefully are in particular need of this feature of MAS: if a single entity controls everything, then the entire system could crash if there is a single failure. Although a multiagent system need not be implemented on multiple processors, to provide full robustness against failure, its agents should be distributed across several machines. Another benefit of multiagent systems is their scalability. Since they are inherently modular, it should be easier to add new agents to a multiagent system than it is to add new capabilities to a monolithic system. Systems whose capabilities and parameters are likely to need to change over time or across agents can also benefit from this advantage of MAS. From a programmer’s perspective the modularity of multiagent systems can lead to simpler programming. Rather than tackling the whole task with a centralized agent, programmers can identify subtasks and assign control of those subtasks to different
5 The Predator/Prey (“Pursuit”)
Domain
The Predator/Prey, or ―Pursuit‖ domain is an appropriate one for illustration of MAS because it has been studied using a wide variety of approaches and because it has many different instantiations that can be used to illustrate different multiagent scenarios. It is not presented as a complex real world domain, but rather as a toy domain that helps concretize many concepts. The pursuit domain was introduced by Benda et al. [2]. Over the years, researchers have studied several variations of its original formulation. The pursuit domain is usually studied with four predators and one prey. Traditionally, the predators are blue and the prey is red. The domain can be varied by using different numbers of predators and prey. The goal of the predators is to ―capture‖ the prey, or surround it so that it cannot move to an unoccupied position. If the world has edges, fewer than four predators can capture the prey by trapping it against an edge or in a corner. Another possible criterion for capture is that a predator occupies the same position as the prey. Typically, however, no two players are allowed to occupy the same position. The size of the world may also vary from an infinite plane to a small, finite board with edges. The predators and prey can move off one end of the board and come back on the other end. Other parameters of the game that must be specified are whether the players move simultaneously or in turns; how much of the world the predators can see; and whether and how the predators can communicate. Finally, in the original formulation of the domain, and in most subsequent studies, the prey moves randomly: on each turn it moves in a random direction, staying still with a certain probability in order to simulate being slower than the predators. However, it is also possible to allow the prey to actively try to escape capture. The parameters that can be varied in the pursuit domain are summarized below. Definition of capture Size and shape of the world Legal moves Simultaneous or sequential movement Visible objects and range Predator communication Prey movement The pursuit domain is a good one for the purposes of illustration because it is simple to understand and because it is flexible enough to illustrate a variety of scenarios. The possible actions of the predators and prey are limited and the goal is well defined. In terms of the reasons to use MAS as presented in the pursuit domain does not necessarily require MAS. But in certain instantiations it can make use of the parallelism, robustness, and simpler programming offered by MAS. In the pursuit domain, a single agent approach is possible: the agent can observe the positions of all four predators and decide how each of them should move. Since the prey moves randomly rather than intentionally, it is not associated with any agent. Instead it is considered part of the environment. It is also possible to consider DPS approaches to the pursuit domain by
breaking the task into sub problems to be solved by each predator. However, most of the approaches described here model the predators as independent agents with a common goal. Thus, they comprise a multiagent system. For each of the multiagent scenarios presented below, a new instantiation of the pursuit domain is defined. Their purpose is to illustrate the different scenarios within a concrete framework.
5.1 Homogeneous Non-Communicating
Multiagent Systems
The simplest multiagent scenario involves homogeneous non-communicating agents. In this scenario, all of the agents have the same internal structure including goals, domain knowledge, and possible actions. They also have the same procedure for selecting among their actions. The only differences among agents are their sensory inputs and the actual actions they take: they are situated differently in the world.
5.1.1 Homogeneous Non-Communicating Multiagent
Pursuit
In the homogeneous non-communicating version of the pursuit domain, rather than having one agent controlling all four predators, there is one identical agent per predator. Although the agents have identical capabilities and decision procedures, they may have limited information about each other’s internal state and sensory inputs. Thus they may not be able to predict each other’s actions. Within this framework, Stephens and Merx propose a simple heuristic behavior for each agent that is based on local information [3]. They define capture positions as the four positions adjacent to the prey. They then propose a ―local‖ strategy whereby each predator agent determines the capture position to which it is closest and moves towards that position. The predators cannot see each other, so they cannot aim at different capture positions. Of course a problem with this heuristic is that two or more predators may move towards the same capture position, blocking each other as they approach. This strategy is not very successful, but it serves as a basis for other control strategies.
5.1.2 General Homogeneous MAS
There are several different agents with identical structure (sensors, effectors, domain knowledge, and decision functions), but they have different sensor input and effector output. That is to say, they are situated differently in the environment and they make their own decisions regarding which actions to take. Having different effector output is a necessary condition for MAS: if the agents all act as a unit, then they are essentially a single agent. In order to realize this difference in output, homogeneous agents must have different sensor input as well. Otherwise they will act identically. For this scenario, in which we consider non-communicating agents, assume that the agents cannot communicate directly.
MAS, but because it was designed for real time control in dynamic environments, it is likely to be extendible to multiagent scenarios.
5.1.4.2 Local or global perspective
Another issue to consider when building a multiagent system is how much sensor information should be available to the agents. Even if it is feasible within the domain to give the agents a global perspectives of the world, it may be more effective to limit them to local views. Roychowdhury et al. consider a case of multiple agents sharing a set of identical resources in which they have to learn (adapt) their resource usage policies [5]. Since the agents are identical and do not communicate, if they all have a global view of the current resource usage, they will all move simultaneously to the most under used resource. However, if they each see a partial picture of the world, then different agents deviates towards different resources: a preferable effect. Better performance by agents with less knowledge is occasionally summarized by the cliche ―Ignorance is Bliss.‖
5.1.4.3 Modeling of other agents’ states
Durfee gives another example of ―Blissful Ignorance,‖ mentioning it explicitly in the title of his paper: ―Blissful Ignorance: Knowing Just Enough to CoordinateWell‖ [13]. Now rather than referring to resource usage, the saying applies to the limited Recursive Modeling Method (RMM). Even if further information can be obtained by reasoning about what agent A thinks agent B thinks agent A thinks, endless reasoning can lead to inaction. Durfee contends that for coordination to be possible, some potential knowledge must be ignored. As well as illustrating this concept in the pursuit domain [14], Durfee goes into more detail and offers more generally applicable methodology in [13]. The point of the RMM is to model the internal state of another agent in order to predict its actions. Even though the agents know each other’s goals and structure (they are homogeneous), they may not know each other’s future actions. The missing pieces of information are the internal states (for deliberative agents) and sensory inputs of the other agents. How and whether to model other agents is a ubiquitous issue in MAS. In the more complex multiagent scenarios presented in the next sections, agents may have to model not only the internal states of other agents, but also their goals, actions, and abilities. Although it may be useful to build models of other agents in the environment, agent modeling is not done universally. Schmidhuber advocates a form of multiagent reinforcement learning (RL) with which agents do not model each other as agents [7]. Instead they consider each other as parts of the environment and affect each other’s policies only as sensed objects. The agents pay attention to the reward they receive using a given policy and checkpoint their policies so they can return to successful ones. Schmidhuber shows that the agents can learn to cooperate without modeling each other.
5.1.4.4 How to affect others
When no communication is possible, system designers must decide how the agents will affect one another. Since they exist in the same environment, the agents can affect each other in several ways. Actively, they can be sensed by other agents, or they may be able to change the state of another agent by, for example, pushing it. More indirectly, agents can affect other agents by one of two types of stigmergy [9]. First, active stigmergy
occurs when an agent alters the environment so as to affect the sensory input of another agent. For example, a robotic agent might leave a marker behind it for other agents to observe. Goldman and Rosenschein demonstrate an effective form of active stigmergy in which agents heuristically alter the environment in order to facilitate future unknown plans of other agents [8].
5.2 Heterogeneous Non-Communicating
Multiagent Systems
To this point, we have only considered agents that are homogeneous. Adding the possibility of heterogeneous agents in a multiagent domain adds a great deal of potential power at the price of added complexity. Agents might be heterogeneous in any of a number of ways, from having different goals to having different domain models and actions. An important sub-dimension of heterogeneous agent systems is whether agents are benevolent or competitive. Even if they have different goals, they may be friendly to each other’s goals or they may actively try to inhibit each other.
5.2.1 Heterogeneous Non-Communicating
Multiagent Pursuit before exploring the general multiagent scenario involving heterogeneous non-communicating agents, consider how this scenario can be instantiated in the pursuit domain. As in the previous scenario, the predators are controlled by separate agents. But they are no longer necessarily identical agents: their goals, actions and domain knowledge may differ. In addition, the prey, this inherently has goals different from those of the predators, can now be modeled as an agent.
5.2.2 General Heterogeneous MAS
As in the homogeneous case the agents are situated differently in the environment which causes them to have different sensory inputs and necessitates their taking different actions. However in this scenario, the agents have much more significant differences. They may have different goals, actions, and/or domain knowledge. This condition of heterogeneity among agents adds a great deal of power for the system designer. In order to focus on the benefits (and complexity) of heterogeneity, the assumption of no communication is retained for this section.
5.2.3 Techniques
The existing techniques in the Heterogeneous non-communicating multiagent system are Game theory, iterative play. Mor and Rosenschein/Sandholm and Crites [15, 16] MinimaxQ Littman [17] Competitive coevolution. Hynes and Sen/Grefenstette and Daley/Rosin and Belew [18] Deduce intentions through observation. Huber and Durfee [19] Autoepistemic reasoning (ignorance). Permpoontanalarp [20] Model as a team (individual role). Tambe [21]
repeated play can cooperation emerge among the selfish agents in the prisoner’s dilemma.
5.2.4.2 Stable vs. evolving agents
Another important characteristic to consider when designing multiagent systems is whether the agents are stable or evolving. Of course evolving agents can be useful in dynamic environments. But particularly when using competitive agents, allowing them to evolve can lead to complications. Such systems that use competitive evolving agents are said to use a technique called competitive co-evolution. Systems that evolve benevolent agents are said to use cooperative co-evolution. The evolution of both predator and prey agents by Haynes and Sen [24] qualifies as competitive co-evolution. Grefenstette and Daley conduct a preliminary study of competitive and cooperative coevolution in a domain that is loosely related to the pursuit domain [25]. Their domain has two robots that can move continuously and one morsel of (stationary) food that appears randomly in the world. In the cooperative task, both robots must be at the food in order to ―capture‖ it. Since the robots can run out of energy if they move too much, they learn to move towards food only when both of them are near enough to reach it. Evolving populations of decision rules using Genetic Algorithms (GAs), Grefenstette and Daley consider different methods of fitness evaluation. Fitness evaluation is an important component of evolutionary learning techniques. Grefenstette and Daley find that an effective method for cooperative coevolution in their domain is to use separate GAs to evolve rules for the two agents, evaluating individuals against a ―champion‖ (individual with highest fitness) from a random generation of the other GA. In a competitive task in the same domain, agents try to be the first to reach the food [25]. Again, different GA evaluation methods are considered for use in evolving rule sets to control the agents. One problem to contend with in competitive rather than cooperative coevolution is the possibility of an escalating ―arms race‖ with no end. Competing agents might continually adapt to each other in more and more specialized ways, never stabilizing at a good behavior. Of course in a dynamic environment, it may not be feasible or even desirable to evolve a stable behavior. Applying RL to the iterated prisoner’s dilemma, Sandholm and Crites find that a learning agent is able to perform optimally against a fixed opponent [16]. But when both agents are learning, there is no stable solution. Another issue in competitive coevolution is the credit/blame assignment problem. When performance of an agent improves, it is not necessarily clear whether the improvement is due to an improvement in that agent’s behavior or a negative change in the opponent’s behavior. Similarly, if an agent’s performance gets worse, the blame or credit could belong to that agent or to the opponent. One way to deal with the credit/blame problem is to fix one agent while evolving the other and then switch. Of course this method encourages the arms race more than ever. Nevertheless, Rosin and Belew use this technique, along with an interesting method for maintaining diversity in genetic populations, to evolve agents that can play TicTacToe, Nim, and a simple version of Go [18]. When it is a given agent’s turn to evolve, it executes a standard GA generation. Individuals are tested against individuals from the competing population, but a technique called ―competitive fitness sharing‖ is used to maintain diversity. When using this technique, individuals from agent X’s population are given more credit for beating opponents (individuals from agent Y’s population) that are not beaten by other
individuals from agent X’s population. More specifically, the reward to an individual for beating individual is divided by the number of other individuals in agent X’s population that also beat individual. Competitive fitness sharing shows much promise for people building systems that use competitive co-evolution.
5.2.4.3 Modeling of others’ goals, actions, and knowledge
In the case of homogeneous agents, it was useful for agents to model the internal states of ther agents in order to predict their actions. With heterogeneous agents, the problem of modeling others is much more complex. Now the goals, actions, and domain knowledge of the other agents may also be unknown and thus need modeling. Without communication, agents are forced to model each other strictly through observation. uber and Durfee consider a case of coordinated motion control among multiple mobile robots under the assumption that communication is prohibitively expensive [19]. Thus the agents try to deduce each other’s plans by observing their actions. In particular, each robot (simulated or real) tries to figure out the destinations of the other robots by watching how they move. Plan recognition of this type is also useful in competitive domains, since knowing an opponent’s goals or intentions can make it significantly easier to defeat. In addition to modeling agents’ goals through observation, it is also possible to learn their actions. Wang’s OBSERVER system allows an agent to incrementally learn the preconditions and effects of planning actions by observing domain experts. After observing for a time, the agent can then experimentally refine its model by practicing the actions itself. When modeling other agents, it may be useful to reason not only about what is true and what is false, but also about what is not known. Such reasoning about ignorances is called autoepistemic reasoning. For a theoretical presentation of an autoepistemic reasoning method in MAS, see [20]. Just asRMMis useful for modeling the states of homogeneous agents, it can be used in the heterogeneous scenario as well. Tambe takes it one step further, studying how agents can learn models of teams of agents. In an air combat domain, agents can use RMM to try to deduce an opponents’ plan based on its observable actions. For example, a fired missile may not be visible, but the observation of a preparatory maneuver commonly used before firing could indicate that a missile has been launched. When teams of agents are involved, the situation becomes more complicated. In this case, an opponent’s actions may not make sense except in the context of a team maneuver. Then the agent’s role within the team must be modeled. Tambe discusses the advantages of team modeling [21]. One reason that modeling other agents might be useful is that agents sometimes depend on each other for achieving their goals. Unlike in game theory where agents can cooperate or not depending on their utility estimation, there may be actions that require cooperation for successful execution. For example, two robots may be needed to successfully push a box, or, as in the pursuit domain, several agents may be needed to capture an opponent. Sichman and Demazeau analyze how the case of conflicting mutual models of different codependent agents can arise and be dealt with [22].
5.2.4.4 Resource management
Heterogeneous agents may have interdependent actions due to limited resources needed by several of the agents. Example domains include network traffic problems in which several different agents must send information through the same network; and load
learn what roles they should fill in different situations. Although already mentioned above in the context of modeling other agents, Tambe’s work deserves mention in this context as well. When an agent is faced with an opposing team of agents, it may be useful to model individual agents as filling roles within a team action rather than as acting independently.
5.3 Heterogeneous Communicating Multiagent
Systems
The scenarios examined thus far have included agents that differ in any number of ways, including their sensory data, their goals, their actions, and their domain knowledge. Such heterogeneous multiagent systems can be very complex and powerful. However the full power of MAS can be realized when adding the ability for agents to communicate with one another. In fact, adding communication introduces the possibility of having a multiagent system turn into a system that is essentially equivalent to a single agent system. By sending their sensor inputs to and receiving their commands from one agent, all the other agents can surrender control to that single agent. In this case, control is no longer distributed. Thus communicating heterogeneous agents can span the full range of complexity in agent systems. Admittedly, communication could be viewed as simply part of an agent’s interaction with the environment. However just as agents are considered special parts of the environment for the purposes of this survey, so is communication among agents considered extra environmental. With the aid of communication, agents can coordinate much more effectively than they have been able to up to this point. In this scenario we include homogeneous as well as heterogeneous communicating agents.
5.3.1 Heterogeneous Communicating Multiagent Pursuit
In the pursuit domain, communication creates new possibilities for predator behavior. Here, agents can still be fully heterogeneous. But now cooperating agents can also communicate with one another. Since the prey acts on its own in the pursuit domain, it has no other agents with which to communicate. However the predators can freely exchange information in order to help them capture the prey more effectively Tan uses communicating agents in the pursuit domain to conduct some interesting multiagent Qlearning experiments [29]. In his instantiation of the domain, there are several prey agents and the predators have limited vision so that they may not always know where the prey are. Thus the predators can help each other by informing each other of their sensory input. Tan shows that they might also help each other by exchanging reinforcement episodes and/or control policies. With communication possible, they define two more possible strategies for the predators [28]. When using a ―distributed‖ strategy, the agents are still homogeneous, but they communicate to insure that each moves toward a different capture position. In particular, the predator farthest from the prey chooses the capture position closest to it, and announces that it will approach that position. Then the next farthest predator chooses the closest capture position from the remaining three, and so on. This simple protocol encourages the predators to close in on the prey from different sides. A distributed strategy, it is much more effective than the local policy and does not require very much communication. However there are situations in which it does not succeed.
Stephens and Merx then present one more strategy that always succeeds but requires much more communication: the ―central‖ strategy [28]. The central strategy is effectively a single agent system. Three predators transmit all of their sensory inputs to one central agent which then decides where all the predators should move and transmits its decision back to them.
5.3.2 General Communicating MAS
Indeed, this continuum of complexity leading into the extreme single agent case applies for MAS in general. With communicating agents, systems can get arbitrarily complex and arbitrarily centralized until a single agent has all the control. Of course communication bandwidth may be prohibitively low to reach the extreme in a given domain. The fully general multiagent scenario appears in Figure 11. In this scenario, we allow the agents to be heterogeneous to any degree from homogeneity to full heterogeneity. The key addition is the ability for agents to transmit information directly to each other. From a practical point of view, the communication might be broadcast or posted on a ―blackboard‖ for all to interpret, or it might be targeted point to point from an agent to another specific agent.
5.3.3 Techniques
The techniques used in the heterogeneous communicating multiagent systems are Speech acts. Cohen and Levesque/Lux and Steiner [31] Bayesian learning in negotiation: model others. Zeng and Sycara [32] Minimize the need for training. Potter and Grefenstette [34] Cooperative co-evolution. Bull et al. [33]
5.3.4 Issues
Since heterogeneous communicating agents can choose not to communicate, and in some cases can also choose to be homogeneous or at least to minimize their heterogeneity, most of the issues discussed in the previous two scenarios apply in this one as well. But the ability to communicate raises another whole set of issues for which techniques exist. Two of the most studied issues are communication protocols and theories of commitment. The issue of benevolence vs. competitiveness, already discussed in the previous MAS scenario, becomes more complicated in this context.
5.3.4.1 Understanding each other
In all communicating multiagent systems, and particularly in domains that include agents built by different designers, there must be some set language and protocol for the agents to use when interacting. Independent aspects of protocols are information content, message format, and coordination conventions. There has been a lot of research done on refining these and other communication protocols. MAS designers must carefully consider what features in a communication protocol are needed in a given domain.
the proper resources) or subcontract them to still other agents. Agents must pay to contract their tasks out and thus shop around for the lowest bidder. Sandholm and Lesser discuss some of the issues that arise in contract nets. In a similar spirit is an implemented multiagent system that controls air temperature in different rooms of a building. A person can set one’s thermostat to any temperature. Then depending on the actual air temperature, the agent for that room tries to ―buy‖ either hot or cold air from another room that has an excess. At the same time, the agent can sell the excess air at the current temperature to other rooms. Modeling the loss of heat in the transfer from one room to another, the agents try to buy and sell at the best possible prices. The market regulates itself to provide equitable usage of a shared resource
5.3.4.4 Resource management
In the previous scenario, resource management came up as a problem involving interdependent actions. In the current scenario, agents can also coordinate schedules. Decker’s Generalized Partial Global Planning (GPGP) allows several heterogeneous agents to post constraints, or commitments to do a task by some time, to each other’s local schedulers and thus coordinate without the aid of any centralized agent.
5.3.4.5 Commitment/decommitment
When agents communicate, they may decide to cooperate on a given task or for a given amount of time. In so doing, they make commitments to each other. Committing to another agent involves agreeing to pursue a given goal, possibly in a given manner; regardless of how much it serves one’s own interests. Commitments can make systems run much more smoothly by providing a way for agents to ―trust‖ each other, yet it is not obvious how to get self interested agents to commit to others in a reasonable way. The theory of commitment and decommitment (when the commitment terminates) has consequently drawn considerable attention. For example, Castelfranchi defines three types of commitment: internal commitment an agent agrees to fill a certain role. Setting an alarm clock is an example of internal commitment to wake up at a certain time. Haddadi discusses commitment states as planning states: potential cooperation, precommitment, and commitment. Agents can then use means ends analysis to plan for goals in terms of commitment opportunities. This work is conducted within a model called Belief/Desire/Intention, or BDI. BDI is a popular technique for modeling other agents. Other agents’ domain knowledge (beliefs) and goals (desires) are modeled as well as their ―intentions,‖ or goals they are currently trying to achieve and the methods by which they are trying to achieve them. Rao and Georgeff use the BDI model to build a system for air traffic control. Finally, groups of agents may decide to commit to each other. Rather than the more usual two agent or all agent commitment scenarios, Zlotkin and Rosenschein study situations in which agents may want to form coalitions. Since this work is conducted in a game theory framework, agents consider the utility of joining a coalition in which they are bound to try to advance the utility of other members in exchange for reciprocal consideration. Shehory and Kraus present a distributed algorithm for task allocation when coalitions are either needed to perform tasks or more efficient that single agents. Sandholm and Lesser use a vehicle routing domain to illustrate a method by which agents can form valuable coalitions when it is intractable to discover the optimal coalitions.
6 References
[1] Keith S. Decker. Task environment centered simulation. In M. Prietula, K. Carley, and L. Gasser, editors, Simulating Organizations: Computational Models of Institutions and Groups. AAAI Press/MIT Press, 1996.
[2] M. Benda, V. Jagannathan, and R. Dodhiawala. On optimal cooperation of knowledge sources an empirical investigation. Technical Report BCS–G2010–28, Boeing Advanced Technology Center, Boeing Computing Services, Seattle, Washington, July 1986.
[3] Peter Stone and Manuela Veloso. Towards collaborative and adversarial learning: A case study in robotic soccer. To appear in International Journal of HumanComputer Systems (IJHCS), 1997.
[4] Tucker Balch and Ronald C. Arkin. Motor schemabased formation control for multiagent robot teams. In Proceedings of the First International Conference on MultiAgent Systems (ICMAS95), pages 10–16, Menlo Park, California, June 1995. AAAI Press.
[5] Shounak Roychowdhury, Neeraj Arora, and Sandip Sen. Effects of local information on group behavior. In Adaptation, Coevolution and Learning in Multiagent Systems: Papers from the 1996 AAAI Spring Symposium, pages 78–83, Menlo Park,CA, March
- AAAI Press. AAAI TechnicalReport SS9601.
[6] Edmund H. Durfee. Blissful ignorance: Knowing just enough to coordinate well. In Proceedings of the First International Conference on MultiAgent Systems (ICMAS95), pages 406–413, Menlo Park, California, June 1995. AAAI Press.
[7] Jurgen Schmidhuber. A general method for multiagent reinforcement learning in unrestricted environments. In Adaptation, Coevolution and Learning in Multiagent Systems: Papers from the 1996 AAAI Spring Symposium, pages 84–87, Menlo Park,CA, March 1996. AAAI Press. AAAI Technical Report SS9601.
[8] Claudia Goldman and Jeffrey Rosenschein. Emergent coordination through the use of cooperative statechanging rules. In Proceedings of the Twelfth National Conference on Artificial Intelligence, pages 408–413, Philadelphia, PA, 1994. Morgan Kaufman.
[9] O.E. Holland. Multiagent systems: Lessons from social insects and collective robotics. In Adaptation,Coevolution and Learning in Multiagent Systems: Papers from the 1996 AAAI Spring Symposium, pages 57–62, Menlo Park,CA, March 1996. AAAI Press. AAAI Technical
[10] Ran Levy and Jeffrey S. Rosenschein. A game theoretic approach to the pursuit problem. In Working Papers of the 11th International Workshop on Distributed Artificial Intelligence, pages 195–213, February 1992.
