Experimental Results from Integrating Planning Systems and Simulation Models David Dyke, Mark Salt, Roberto Desimone Peter Jarvis = DERA Malvern, AI Applications Institute St Andrews Road 80 South Bridge, Malvern WR14 3PS, UK Edinburgh EH1 1HN, UK Tel: +44 1684 895447 Tel: +44 131 650 2732 dsdyke@dera.gov.uk Peter.Jarvis@ed.ac.uk Abstract This paper presents results from the implementation of a prototype based on the architecture for integrating planning and simulators proposed last year. The paper describes exactly what has been implemented and uses a non-combatant evacuation scenario (NEO) to show the benefits of such integration. Also discussed are the lessons learned from the results obtained and, finally, a description of a revised and improved architecture, which is believed will address some of the outstanding research issues. 1. Introduction Large planning problems demand the co-operation and collaboration of many disparate and geographically dispersed experts. As part of the planning process, it is desirable to ‘play out’ fragments of the evolving plan, obtain and evaluate feedback, and compare alternative courses of action (COA). One way of achieving this is to use simulators to exercise the plan and provide feedback on the planning process by incorporating measures of effectiveness into plans. However, existing planning systems do not support collaborative planning well and have poor (or no) integration with simulation models and other services. In addition, simulators have been mostly developed in isolation from other systems. This paper reports on continuing research within the UK Defence Evaluation and Research Agency (DERA) with the following key objectives: • To explore the integration of planning systems with simulation models through the use of plan and simulation query languages providing guidance and validation of planning choices. • To demonstrate the benefits of such integration in terms of more effective plans with greater accuracy, robustness and flexibility, hence accelerating the overall planning cycle. • To make recommendations on improvements to description languages for plans and simulation models to better facilitate their integration. Above all the integrated architecture should provide direct links between planning and command decisions made by military planners and validated simulation models that characterise and justify the military effectiveness for each decision, rather than just for the operations plan as a whole. = Dr Jarvis now works at SRI International, Menlo Park, California