Decentralised task allocation using GDL negotiations in Multi-agent system

In large distributed systems, the optimization algorithm of task scheduling may not meet the special requirements of the domain control mechanism, i.e. robustness, optimality, timeliness of solution and computational ease of processing under limited communication. In or- der to satisfy these requirements, a novel decentralized agent scheduling method for dynamic task allocation problems based on Game Descrip- tion Language (GDL) and Game Theory is proposed. Specifically, we define the task allocation problem as a stochastic game model, in which the agent's utility is derived from the marginal utility, and then prove that the global optimal task allocation scheme resides in the Nash equi- librium set by the non-cooperative game. In order to generate an optimal solution, we define Multi-agent Negotiation Game (MNG), in which ne- gotiations are held between agents to decide which tasks to act on next. Building on this, we make a simple extension to adopt GDL more suit- able for negotiations and propose to use it to model such negotiation scenarios. Finally, we use a negotiation example to show that our ap- proach is more amenable to automatic processing by autonomous agents and of great practicality than a centralized task scheduler.