Demonstration of DEFACTO: Training Tool for Incident Commanders Nathan Schurr∗ , Janusz Marecki∗ , Milind Tambe∗ and Paul Scerri+ ∗
1.
University of Southern California, {schurr, marecki, tambe}@usc.edu + Carnegie Mellon University,
[email protected]
OVERVIEW
example, during one of the demonstrations on Nov 18, 2004 Gary Ackerman, a Senior Research Associate at the Center for Nonproliferation Studies at the Monterey Institute of International Studies pointed out in reference to DEFACTO, “This is exactly the type of system we are looking for” to study the potential effect of terrorist attacks. Also, we have met with several public safety officials about using DEFACTO as a training tool for their staff. According to Los Angeles County Fire Department Fire Captain Michael Lewis: “Effective simulation programs for firefighters must be realistic, relevant in scope, and imitate the communication challenges on the fire ground. DEFACTO focuses on these very issues.”
In the wake of large-scale national and international terrorist incidents, it is critical to provide first responders and rescue personnel with tools and techniques that will enable them to evaluate response readiness and tactics, measure inter-agency coordination and improve training and decision making capability. We focus in particular on building tools for training and tactics evaluation for incident commanders, who are in charge of managing teams of fire fighters at critical incidents. Such tools would provide intelligent software agents that simulate first responder tactics, decisions, and behaviors in simulated urban areas and allow the incident commander (human) to interact. These agents form teams, where each agent simulates a fire engine, which plans and acts autonomously in a simulated environment. Through interactions with these software agents, an incident commander can evaluate tactics and realize the consequences of key decisions, while responding to such disasters. To this end, we have constructed a new system in close cooperation with the Los Angeles fire department and and personnel from the Homeland Security Center for Risk and Economic Analysis of Terrorism Events (CREATE). We refer to this system as DEFACTO (Demonstrating Effective Flexible Agent Coordination of Teams through Omnipresence). DEFACTO incorporates state of the art artificial intelligence, 3D visualization and human-interaction reasoning into a unique high fidelity system for training incident commanders. By providing the incident commander interaction with the coordinating agent team in a complex environment, the commander can gain experience and draw valuable lessons that will be applicable in the real world. With DEFACTO, our objective is to both enable the human to have a clear idea of the team’s state and improve agent-human team performance. We want DEFACTO agent-human teams to better prepare firefighters for current human-only teams and better prepare them for the future, in which agent-human teams will be inevitable and invaluable. DEFACTO system has been periodically demonstrated to key government agencies, public safety officials, and disaster researchers for assessing its utility by the ultimate consumers of the technology, with exciting feedback. Indeed, they were eager to deploy DEFACTO and begin using it as a research tool to explore the unfolding of different disasters. For
DEFACTO Incident Commander
Omni-Viewer Navigation
Allocation
Disaster Scenario Team Proxy Proxy Proxy Proxy
Figure 1: DEFACTO system architecture.
2.
KEY FEATURES
Our demonstration of DEFACTO will highlight three key features, namely, the Omni-Viewer, the proxy based coordination and team level adjustable autonomy. Below, we provide a brief overview of each of these features.
2.1
Omni-Viewer
Our goal of allowing fluid human interaction with agents requires a visualization system that provides the human with a global view of agent activity as well as shows the local view of a particular agent when needed. Hence, we have developed an omnipresent viewer, or Omni-Viewer, which will allow the human user diverse interaction with remote agent teams. The Omni-Viewer incorporates both a conventional map-like 2D view, Allocation Mode (Figure 2-d), and a detailed 3D viewer, Navigation Mode (Figure 2b). The Allocation mode shows the global overview as events are progressing and provides a list of tasks that the agents have transfered to the human. The Navigation mode shows the same dynamic world view, but allows for more freedom to move to desired locations and views. In particular, the user can drop to the virtual ground level, thereby obtaining the world view (local perspective) of a particular agent.
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. AAMAS’05, July 25-29, 2005, Utrecht, Netherlands. Copyright 2005 ACM 1-59593-150-2/05/0007 ...$5.00.
147
3.
Among the current tools aimed at simulating rescue environments are JCATS [10] and EPICS [5]. JCATS is a self-contained, high-resolution joint simulation in use for entity-level training in open, urban and subterranean environments. JCATS gives users the capability to replicate in detail activities of smalls group and individuals during a simulated operation. EPICS is a computerbased, scenario-driven, high-resolution simulation. It is used by emergency response agencies to train for emergency situations that require multi-echelon and/or inter-agency communication and coordination. EPICS is also used for exercising communications and command and control procedures at multiple levels. However, neither JCATS nor EPICS allow the simulation of agents, nor do they provide the intelligent interaction capabilities described here. Much other work has looked at how to build intelligent agents for simulation environments. For example, Hill et al’s work is a similar immersive training tool to the one demonstrated here[4]. Their work focused more on multi-modal dialog and emphasize single agent interaction along predefined story lines, whereas this work focuses on adjustable autonomy and coordinating large numbers of agents in a dynamic, complex fire-fighting domain. Human interactions in agent teams has also been investigated in [1, 9], and there is significant research on human interactions with robot-teams [3, 2]. However, they do not use flexible AA strategies and/or team-level AA strategies. Furthermore, the experimental results here may assist these researchers in recognizing the potential for harm that humans may cause to agent or robot team performance.
(a) Figure 2: Omni-Viewer during a scenario: (a) The Incident Commander uses the Navigation mode to quickly grasp the situation (b) Allocation mode is used to assign a fire engine to the fire.
2.2
Proxy: Team Coordination
A key hypothesis in this work is that intelligent distributed agents will be critical to disaster response training. Specifically, coordination algorithms inspired by theories of teamwork will be used to manage the distributed response [8]. Our general coordination algorithms are encapsulated in proxies, with each team member having its own proxy representing it in the team. The current version of the proxies is called Machinetta [7] and extends the successful Teamcore proxies [6]. Machinetta is implemented in Java and is freely available on the web. Notice that the concept of a reusable proxy differs from many other “multiagent toolkits” in that it provides the coordination algorithms, e.g., algorithms for allocating tasks, as opposed to the infrastructure, e.g., APIs for reliable communication.
2.3
RELATED WORK
Acknowledgements This research was supported by the United States Department of Homeland Security through the Center for Risk and Economic Analysis of Terrorism Events (CREATE). However, any opinions, findings, and conclusions or recommendations in this document are those of the author and do not necessarily reflect views of the U.S. Department of Homeland Security.
Proxy: Adjustable Autonomy
Adjustable autonomy refers to an agent’s ability to dynamically change its own autonomy, possibly to transfer control over a decision to a human. A the context of single-agent to single-human interactions, pre-planned sequences of transfer-of-control actions, called transfer-of-control strategies have shown significant promise. For example, an AH1 H2 strategy implies that an agent (A) attempts a decision and if the agent fails in the decision then the control over the decision is passed to a human H1 , and then if H1 cannot reach a decision, then the control is passed to H2 . The notion of flexible strategies, however, has not been applied in the context of humans interacting with agent-teams. Thus, a key question is whether such flexible transfer of control strategies are relevant in agent-teams, particularly in a large-scale application such as ours. DEFACTO aims to answer this question by implementing transfer-of-control strategies in the context of agent teams. One key advance in DEFACTO is that the strategies are not limited to individual agent strategies, but also enable team-level strategies. For example, rather than transferring control from a human to a single agent, a team-level strategy could transfer control from a human to an agent-team. Concretely, each proxy is provided with all strategy options; the key is to select the right strategy given the situation. An example of a team level strategy would combine AT Strategy and H Strategy in order to make AT H Strategy. The default team strategy, AT , keeps control over a decision with the agent team for the entire duration of the decision. The H strategy always immediately transfers control to the human. AT H strategy is the conjunction of team level AT strategy with H strategy. This strategy aims to significantly reduce the burden on the user by allowing the decision to first pass through all agents before finally going to the user, if the agent team fails to reach a decision.
4.[1] Mark REFERENCES H. Burstein, Alice M. Mulvehill, and Stephen Deutsch. An [2] [3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
148
approach to mixed-initiative management of heterogeneous software agent teams. In HICSS, IEEE Computer Society, 1999. Jacob W. Crandall, Curtis W. Nielsen, and Michael A. Goodrich. Towards predicting robot team performance. In SMC, 2003. Terrence Fong, Charles Thorpe, and Charles Baur. Multi-robot remote driving with collaborative control. IEEE Transactions on Industrial Electronics, 2002. R. Hill, J. Gratch, W. L. Johnson, C. Kyriakakis, C. LaBore, R. Lindheim, S. Marsella, D. Miraglia, B. Moore, J. Morie, J. Rickel, M. Thiébaux, L. Tuch, R. Whitney, J. Douglas, and W. Swartout. Toward the holodeck: integrating graphics, sound, character and story. In AGENTS ’01, ACM Press, 2001. Lawrence Livermore National Laboratory. Jcats - joint conflict and tactical simulation. In http://www.jfcom.mil/about/fact jcats.htm, 2005. D. V. Pynadath and M. Tambe. Automated teamwork among heterogeneous software agents and humans. JAAMAS, 7:71–100, 2003. P. Scerri, E. Liao, Yang. Xu, M. Lewis, G. Lai, and K. Sycara. Theory and Algorithms for Cooperative Systems, chapter Coordinating very large groups of wide area search munitions. World Scientific Publishing, 2004. P. Scerri, D. Pynadath, and M. Tambe. Towards adjustable autonomy for the real world. Journal of Artificial Intelligence Research, 17:171–228, 2002. Ulrich Neumann Suya You, Jinhui Hu and Pamela Fox. Urban site modeling from lidar. In Proc. 2nd Int’l Workshop Computer Graphics and Geometric Modeling , 2003. Advanced Systems Technology. Epics - emergency preparedness incident commander simulation. In http://epics.astcorp.com, 2005.
