Don Perugini

A distributed agent approach to global transportation scheduling

A suitable method for distributed transportation scheduling is necessary for the development of an agent-based logistics support system for the Australian Defence Force. Manager Agents (MA) require resources to be transported, in large quantities and on a global scale, using the services of many transport agents (TA). We investigate the application of K. Fischer et al.s (1996) Extended Contract Net Protocol (ECNP) to our transportation domain. ECNP provides a distributed and reactive approach to transportation scheduling that is fairly suited to this domain, but has its shortfalls. This includes the inability for TA to bid for partial routes, in addition to partial quantities, for a transportation task. A new protocol is proposed, based on ECNP, called the Provisional Agreement Protocol (PAP), which overcomes these shortfalls. Five speech acts, together with policies regarding the persistence and commitment of tasks and bids, are proposed. This allows agents to revisit bid options that were previously selected, and thus accomodates bidding for partial routes.

Distributed information fusion agents

Information fusion is typically data driven and applied to adversarial contexts. Advantages of a goal driven fusion process are emerging. An agent approach to fusion applied to friendly contexts, such as logistics, is presented. Agents are suitable for fusion because they can represent autonomous fusion eritities by modelling their capabilities, expertise and intentions. This paper promotes a level 3 centric view of information fusion, and focuses on impact and situation fusion. Extended Contract Net Protocol (ECNP) provides a distributed fusion approach, but has its shortfalls for the fusion domain. Provisional Agreement Prorocol (PAP) is an extension of the ECNP. It allows backtracking and a deliberative approach to fusion. Details of PAP are presented.

Provisional agreement protocol for global transportation scheduling

The global transportation scheduling problem is complex, decentralised, open and dynamic. It typically requires the services of many transport organizations to transport partial quantities along partial routes to fulfill a transportation task. We have applied agents to address this problem. The Provisional Agreement Protocol (PAP) was developed to facilitate the planning required in our transportation problem. A greedy PAP approach has been implemented for the complex global transportation problem, allowing partial quantity and route bids, and backtracking if an infeasible solution is encountered. In this paper, we present the PAP, together with some improvements over that which has been previously presented. Further implementation details and formal evaluation are provided. Our implementation allows a wider range of transportation problems to be solved than previous approaches.

From single static to multiple dynamic combinatorial auctions

We apply the provisional agreement protocol (PAP) as a new approach to single static, single dynamic and multiple combinatorial auction problems, and empirically evaluate PAP. PAP benefits over one-shot auctions include: bidders not required to submit all bids and their dependencies; interaction with a changing environment during the auction can improve the solution; less communication when each bidder possesses many bids. PAPs backtracking may allow a better solution to be found than the first (greedy) solution, but can be detrimental with multiple auctions when bids (resources) are limited. With multiple auctions, dynamics and competition increases as resources becomes scarce. Therefore, PAP is likely to perform better when many resources are available, which is when auctions are useful anyway. PAP scales well, and applying PAP to a second domain shows its generality.

Agent-Based Global Transportation Scheduling in Military Logistics

Provisional Agreement Protocol (PAP) facilitates decentralised agent planning in open and dynamic domains. PAP is implemented for our global transportation scheduling problem, allowing partial quantity and route bids, and backtracking if an infeasible solution is encountered. Current experimental results are presented, which show that our agent-based transportation scheduling implementation performs well compared with finding transportation schedules manually, which is how military logistics planners form schedules in training.

Dynamic agent systems in the CoAX Binni 2002 experiment

The goal of the intemational CoAX (Coalition Agents experiment) program was to demonstrate how agent systems could be used to provide agile andflexible command and control systems for coalition operations, and facilitate rapid integration of national C2 systems. The CoAX experiments modelled a coalition C4ISR system as a distributed, heterogeneous agent network using the DARPA CoABS (Control of Agent Based Systems) Grid infrastructure based on Java JINI technology. This paper outlines the CoAX Binni experiment which was held in October 2002 at the US Naval Warfare College, Newport RI. It describes the technology used in this experiment and the role of the ATI-ITUDE multi-agent architecture in the Australian componenr of the eqeriment. This involved logistics planning (and dynamic replanning) for a casualty evacuation from an Australian ship using BDI agents developed in the A~ITVDE architecture, and included interactions with Coalition medical and planning agents. Distributed agents were used to represent the various organisational entities involved in a simpl

Order of Movement in Decommitting from a Contract

ed logistics model, and agent interactions with the Coalition C4ISR system were mediated by human operators using I-X Process Panels. This provided a semi-autonomous system, where human approval initiated further autonomous interactions between Coalition and Australian agents.

A Distributed Agent Approachto Global Transportation Scheduling

There is increasing interest in automated agent negotiation and contracting in both commercial and military applications. Sandholm and Lesser proposed leveled commitment contracts allowing agents to decommit from contracts by paying a penalty. Further to this study we analyse the order of movement (decision) in contract decommitment. Intuitively one would expect that moving second is better. We show that this is not always the case. Agents may prefer to move first and not decommit in order to force the other agent to decommit and pay the penalty. Electing to move second may reveal information about an agents type, altering its expected payoff and cause instability with the Nash equilibrium presented. This may result in agents that prefer to move second or are indifferent to order to move first or second (or use a mixed strategy) respectively. If at least one agent must pay a large penalty (i.e. full commitment contract), both agents are indifferent to when they move.