Decentralized Decision Making in Multi-Agent Systems

Abstract

The literature in the feld of multi-agent manufacturing control predom-inantly presents qualitative arguments to motivate its usage. This workpresents quantitative meaning to some of these arguments. The work alsodiscusses the control structuring process and puts it into perspective withmulti-agent control. A suitable example for the analysis of multi-agent con-trol is developed. A two layered approach for the control of this exampleprocess is proposed and it is shown that agents can be used in both of theselayers. An ad-hoc analysis is made of the top layer and simulation resultsshows that a distributed control approach introduces an optimality gap whencompared to a centralized approach. Further simulation results show thata layered approach to multi-agent control for the proposed process improvesoverall performance by increasing agent coordination.To date there has been a lack of industry adoption of multi-agent tech-nologies. The industry seems to favor control approaches that are well testedand with quantifiable evidence supporting their efficiency. The developmentof simulators that capture the full complexity of manufacturing processes isa time consuming task. The example process proposed in this work onlycaptures some core dynamics. Further work on presenting more quantifiablearguments for the use of multi-agent control is thus motivated throughoutthe thesis.The ideas presented in this thesis are included in the article MAS formanufacturing control: A layered case study by Pedersen et al. The articleis aimed for submission to the AAMAS 2012 conference in Valencia.