MADeM: Multi-modal Agent Decision Making

MADeM overview

The MADeM (Multi-modal Agent Decision Making) model provides agents with a general mechanism to make socially acceptable decisions. In this kind of decisions, the members of an organization are required to express their preferences with regard to the different solutions for a specific decision problem. The whole model is based on the MARA (Multi-Agent Resource Allocation) theory, therefore, it represents each one of these solutions as a set of resource allocations. MADeM can consider both tasks and objects as plausible resources to be allocated, which it generalizes under the term task-slots. MADeM uses first-sealed one-round auctions as the allocation procedure and a multi-criteria winner determination problem to merge the different preferences being collected according to the kind of agent or society simulated. Thus, the formal definition of a MADeM decision can be represented by the following tuple:

<a, Al, Ag, Pw, Uf, Uw, Cuf>

where:

a ∈ A is the agent in charge of making a social decision, where A is the set of all agents in the society.

Al is the set of resource or task-slots allocations representing all possible solutions for a specific decision problem.

Ag ⊂ A is the subset of agents being consulted or target agents, which can be either infered from the organisational structure or maintained locally.

Pw: Ag → ℜ are the personal weights (i.e. personal attitudes) that are used to balance the preferences received from each agent in Ag.

Uf is the set of utility functions of the form u: Al x Ag → ℜ representing the agents' preferences with regard to the resource allocations considered.

Uw: Uf → ℜ are the utility weights that are used by the agent a to balance the importance given to each utility function in Uf when resolving the winner determination problem.

Cuf ∈ {elitist, egalitarian, utilitarian, nash} is the collective utility function representing the social welfare of the simulated society, that is, the type of society where agents are located.

For a complete explanation of the MADeM model click here.

J-MADeM

MADeM is available as an open source library fully integrated into Jason under the name J-MADeM. The aim of J-MADeM is to improve Jason by incorporating the main features of MADeM. Thus, J-MADeM agents can easily define the utility functions expressing their preferences and find socially acceptable decisions for specific decision problems.

J-MADeM examples

  • Gold Miners
      J-MADeM Gold Miners  

    This example shows how to easily integrate J-MADeM in a well known Jason example: the Gold Miners.

    Here, miner agents use the MADeM process to decide the best miner to allocate the gold to. The whole process (launching the auction, getting bids and calculating winner) is performed inside the miner who started the MADeM process. Thus, the leader agent is not in charge of getting the bids and allocating the gold any more.

    In order to calculate their bids, miners use the utility function goldDistance. This function computes the preference towards a gold according to the distance between the agent and the gold location, so that the gold can be allocated to the nearest miner.
  • Gold Miners Structured
      Gold Miners Structured  

    This example shows how to easily create a dynamic multi-agent organization that uses J-MADeM in a well known Jason example: the Gold Miners.

    Here, we use a multi-agent structure of miners and bosses to adapt better to nonuniform gold distributions. On the one hand, miner agents just inform their corresponding boss when they find a chunk of gold. Although bosses are not allowed to directly pick up pieces of gold, they can allocate them to its subordinated miners.

    Initially a balanced miner-boss assignment is performed, so that all bosses manage the same number of miners. However, they can dynamically change the organization by modifying this miner-boss relations during the simulation. This way, bosses are allowed to borrow miners when the number of gold chunks found in the quadrant they control increases.
  • Preference meetings
      Performance of Meeting Scheduling Problem  

    This example presents a J-MADeM implementation of the Meeting Scheduling Problem, a well known scheduling problem to find the best allocation for a meeting within an organization.

    The example proposed is about school meetings, where three agent roles (i.e. director, teacher, father and worker) try to schedule some meetings. Two kinds of meetings are considered: Usual monomodal meetings, based on laboral preferences and MADeM Multimodal meetings considering laboral and personal preferences.

    The average number of attendants is higher when applying multimodality, as it offers intermediate solutions between both laboral and personal points of view.
  • Urban mobility
      Urban mobility  

    This example corresponds to an urban mobility simulation framework developed over Jason that allows to model the mobility within a metropolitan area.

    J-MADeM has been used in this scenario to model the main decision that inhabitants make every morning. That is, which transport to use for traveling to work: alone in their own car, sharing a car or by train. Therefore, the proposed approach focuses on the decision making aspects of this problem at a micro level, instead of focusing on the classical spatial or other macro level issues.

    The first results show the behavior of two societies of individualist and egalitarian citizens, which affect the average travel time, the use of the urban transportation and the amount of CO2 emitted to the environment.
  • Virtual University Bar
      Snapshot of the university bar  

    This social scenario represents a virtual university bar where waiters take orders placed by customers. Both of them use J-MADeM to decide among different task assignments.

    In this example, we tested the ability of J-MADeM to model different social welfares (e.g. elitism and utilitarianism) as well as personal attitudes (e.g. altruism, egoism, reciprocity...). For a full description of the simulated scenario and the social outcomes produced click here

    The following video shows a running where a group of waiters display social behaviors such as task passing, reciprocity and planned meetings. In turn, customers try to share table with those belonging to the same social class (represented as different avatars).

Related publications

Some documents distributed by this server have been provided by the contributing authors as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.

Towards a Model for Urban Mobility Social Simulation - A perspective from J-MADeM Decision Making. Francisco Grimaldo, Miguel Lozano, Fernando Barber, Alejandro Guerra-Hernández. Special Issue on MAS Applications in Transports. Journal Progress in Artificial Intelligence. May 2012.
doi: http://dx.doi.org/10.1007/s13748-012-0012-z.

J-MADeM, a market-based model for complex decision problems Francisco Grimaldo, Miguel Lozano, Fernando Barber. Logic Journal of the IGPL. Oxford University Press. April 2011.
doi: http://dx.doi.org/10.1093/jigpal/jzq028.

J-MADeM v1.1: A full-fledge AgentSpeak(L) multimodal social decision library in Jason Francisco Grimaldo, Miguel Lozano, Fernando Barber, Alejandro Guerra-Hernández. The 8th European Workshop on Multi-Agent Systems (EUMAS 10). Paris (France), December 2010.

J-MADeM, an open-source library for social decision-making Francisco Grimaldo, Miguel Lozano, Fernando Barber. 12è Congrés Internacional de l'Associació Catalana d'Intel·ligència Artificial (CCIA 2009). Cardona (Catalonia-Spain), October 2009.

MADeM: a multi-modal decision making for social MAS Francisco Grimaldo, Miguel Lozano, Fernando Barber. Autonomous Agents and Multiagent Systems (AAMAS 2008). ACM Press, Estoril (Portugal), May 2008.