On-line Planning Simulation

MESS is a substrate for building simulation environments suitable for testing plans and on-line or realtime planners. The article describes the design of MEss, how simulations are built and how on-line planners integrate with the substrate. MEss supports activities, defined as processes over some time interval, and interactions between activities and other simulation events. MEss interfaces with TCL, which is a portable, extensible definition of computation time, enabling MESS to be used for platform-independent simulations of real-time planning. MESS has been used to re-implement the PHOENIX testbed, which simulates forest fires and planning for fire-fighting agents. The Need for Simulation As planners become more sophisticated, they will solve increasingly large planning problems involving, for example, the movement and actions of thousands of vehicles, over many hours and under changing conditions. It is extremely difficult to inspect such elaborate plans and determine, for example, their probability of success, the extent to which their goals will be satisfied, and so forth. Nevertheless, such evaluation is critical to a scientific understanding of how and how well a sophisticated planner works. We believe simulation is necessary to evaluate planners: plans are run man)" times in the space of conditions that they were meant to handle, and various dependent variables are measured and statistically analyzed. Furthermore, simulators enable the planner to be on-line: it can be an agent in an ongoing environment, monitoring the progress of the plan and making additions or corrections as necessary. An on-line planner can even scrap a failing plan or sub-plan and replan (Howe 1993). If the thinking time of the planner is limited, so that there is time pressure on its thinking, the on-line planning becomes real-time planning. A number of simulation environments already exist to support research in on-line and real-time planning (Hanks, Pollack, & Cohen 1993). Some of these simulators are quite domain-specific, such as our own PHOENIX testbed (Cohen et al. 1989), which simulates forest fires in Yellowstone National Park. Other examples are TRUCKWORLD (Hanks, Nguyen, & Thomas 1992) and TRAINS (Martin & Mitchell 1994), where trucks or trains move cargo in a graph of depots, cities and towns. Other testbeds are much more domainindependent, such as the MICE testbed (Durfee Montgomery 1990), in which agents move in a generic gridworld. With such a plethora of testbeds, there have been many good ideas and much duplication of effort. In addition to implementing the domain dynamics, these testbeds all have to solve the fundamental issues of simulation, such as managing events from many sources and getting them to occur in the correct order. They have to deal with the interface between planners and the environment, and often that interface is not well defined. Will the testbed have multiple agents, and how is their concurrent thinking coordinated? How is thinking time represented and integrated into a discrete event simulation? The solutions are often not as flexible and powerful as the planning community might like. Because of these many design decisions, these testbeds are often not as easily shared as their authors intended. This article describes our work on MESS (Anderson 1995), which we believe captures the best of the common, domain-independent aspects of these simulators, and improves the representation of thinking agents and the measurement of time. MESS (Multiple Event Stream Simulator) is best described as a simulation substrate, rather than a simulated environment in itself. It makes no domain commitment because it works with abstractions called "events," "event streams" and "activities," among others. One builds a simulation environment in MESS by defining the events that happen, thereby changing the state of the world, and defining the event streams that produce those events. The MESS substrate takes care of synchronizing all the events so that the simulation Anderson 3 From: AIPS 1996 Proceedings. Copyright © 1996, AAAI (www.aaai.org). All rights reserved.