Logo image
Reasoning intra-dependency in commitments for robust scheduling
Conference paper   Peer reviewed

Reasoning intra-dependency in commitments for robust scheduling

Mingzhong Wang, K Ramamohanarao and J Chen
Proceedings of the 8th International Conference on Autonomous Agents and Multiagent Systems, Vol.1, pp.718-725
International Conference on Autonomous Agents and Multiagent Systems (AAMAS), 8th (Budapest, Hungary, 10-May-2009–15-May-2009)
International Foundation for Autonomous Agents and Multiagent Systems (IFAAMAS)
2009
url
http://www.ifaamas.org/Proceedings/aamas09/pdf/01_Full%20Papers/17_95_FP_0171.pdfView
Webpage

Abstract

commitment machines commitment refactoring agent interaction robustness scheduling
Commitment-modeled protocols enable flexible and robust interactions among agents. However, existing work has focused on features and capabilities of protocols without considering the active role of agents in them. Therefore, in this paper we propose to augment agents with the ability of reasoning about and manipulating their commitments to maximize the system utility. We adopt a bottom-up approach by first investigating the intra-dependency between each commitment's preconditions and result which leads to a novel classification of commitments as well as a formalism to express various types of complex commitment. Within this framework, we provide a set of inference rules to benefit an agent by means of commitment refactoring which enables composition and/or decomposition of its commitments to optimize runtime performance. We also discuss the pros and cons of an agent scheduling and executing its commitments in parallel. We propose a reasoning strategy and an algorithm to minimize possible loss when the commitment is broken and maximize the overall system robustness and performance. Experiments show that concurrent schedules based on the features of commitments can boost the system performance significantly. Copyright © 2009, International Foundation for Autonomous Agents and Multiagent Systems.

Details

Metrics

495 Record Views
Logo image