Um protocolo para contratação de agentes em grupos de larga escala

• Ano de defesa: 2007
• Autor(a) principal: Christian Jorge Delgado Polar
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: agentes; Agentes inteligentes (Software); Inteligencia artificial; Computação; Engenharia de software; Locação de serviços Automação; Contrato de trabalho Automação
• Link de acesso: http://hdl.handle.net/1843/RVMR-794PY6

Multi-Agent Systems have been used in the solution of variousproblems in different fields, ranging from electronic commerce tomobile robotics. In these systems, individual agents may have tocooperate in order to fulfill their objectives. A common way ofcooperation is the execution of contracts, through which an agent maycontract a more capacitated one to execute a specific task. Contractsare particularly important when agents have different objectives. Inthis case, a agent will contract another one that, despite havingother objectives, will execute the task receiving a payment for that.With the advances in massive computing systems, there is anincreasing need for technologies that deal with very large numbers ofagents. In these systems, generally called swarms, the contractingmust be done in a scalable way. The present work presents a scalableprotocol for contracting in large groups of self-interested agents.The protocol has three basic phases: a discovery phase that searchesfor adequate agents for contracting, a negotiation phase, in whichthe contracting price is determined and an execution phase, in whichtasks are allocated and agents receive their payment after taskexecution. The use of dynamic thresholds facilitates agent discovery and a mix of two types of auctions allows agents to negotiate and determinethe values to be paid under the contract. The protocol was tested insimulations where teams of robotic vehicles belonging to differentcompanies must transport items from specific targets. Resultsdemonstrate that the proposed protocol is rational, pareto--efficient, distributed and fair, desirable characteristics in any contracting protocol. Moreover, the protocol is scalable and uses low levels of communication and processing, making it adequate for swarms.