Negotiation mechanisms for the multi-agent multi-mode resource investment problem

Abstract

Carrying out complex projects often involves several collaborating parties (agents) with conflicting goals. We consider project scheduling problems, where each activity belongs to one of several agents, with a given deadline on the project completion while one aims at a schedule with an efficient use of resources. As there are precedence relations among activities and the activity execution requires varying amounts of resources, the need for coordination among the agents arises. For the execution of activities, an agent can choose from several modes that determine the processing time and resource usage. Each agent aims to minimise his/her individual resource costs associated with the project. Hence, the problem at hand is the multi-agent generalisation of the multi-mode resource investment problem. Here, we consider local as well as global resources. The latter ones are shared among the agents, thus the problem involves the need for a suitable allocation of respective resource costs. As agents are unwilling to share critical and sensitive information (or are not expected to always provide truthful information), the employed solution procedure should not rely on such information under consideration of the agents’ incentives. We propose and extend such decentralised negotiation mechanisms which facilitate the allocation of global resources. We analyse their potential to overcome information asymmetry and yield high quality solutions utilising a distributed scheduling procedure and a representation which aids in learning effective mode decisions.