James H. Lawton

Knowledge systems for coalition operations

Most major military, peacekeeping, and humanitarian operations are now coalition-based and require agility and effective use of limited resources to achieve complex and multiple objectives. This raises many challenges given technical incompatibilities, rules and regulations, as well as cultural norms. This special issue examines the contributions of intelligent systems to help address this important problem domain.

Multi-agent Planning in Sokoban

The issue of multi-agent planning in highly dynamic environments is a major impediment to conventional planning solutions. Plan repair and replanning solutions alike have difficulty adapting to frequently changing environment states. To adequately handle such situations, this paper instead focuses on preserving individual agent plans through multi-agent coordination techniques. We describe a reactive agent system architecture in which the main focus of an agent is to be able to achieve its subgoals without interfering with any other agent. The system is a 3-level architecture, where each level is guided by the following fundamental principles, respectively: whenis it valid to generate a plan for a subgoal, whois most appropriate for completing the subgoal, and howshould the plan be carried out.

A Unified Long-Term Memory System

Memory-based reasoning systems are a class of reasoners that derive solutions to new problems based on past experiences. Such reasoners use a long-term memory (LTM) to act as a knowledge base of these past experiences, which may be represented by such things as specific events (i.e. cases), plans, scripts, etc. This paper describes a Unified Long-Term Memory (ULTM) system, which is a dynamic, conceptual memory that was designed to be a general LTM capable of simultaneously supporting multiple intentional reasoning systems. Through a unique mixture of content-independent and domain-specific mechanisms, the ULTM is able to flexibly provide reasoners accurate and timely storage and recall of episodic memory structures. In addition, the ULTM provides support for recognizing opportunities to satisfy suspended goals, allowing reasoning systems to better cope with the unpredictability of dynamic real-world domains by helping them take advantage of unexpected events.

Multi-agent opportunistic planning and plan execution

Multiagent opportunism refers to the ability of agents operating in a multiagent system (MAS) to recognize and respond to potential opportunities for mutual assistance in achieving individual goals. Two major potential obstacles in operationalizing multiagent opportunistic assistance in real-world systems are (i) low amounts of knowledge shared between the agents, and (ii) limited ability of the agents to re-plan dynamically. We have previously shown that even under these limiting conditions, systems of agents can benefit from multiagent opportunism. In this work we discuss how multi-agent systems can exploit shared knowledge for opportunistic predictive encoding using an approach based on an abstract plan representation called partial order plan graphs (POPGs). Further, we present several approaches for increasing system-level performance by improving the efficiency of the plans containing predictively encoded opportunities, as well as the results of an empirical analysis of their impact on the system performance.

On the Role of Knowledge in Multi-Agent Opportunism

Multi-agent opportunism refers to the ability of agents operating in a multi-agent system (MAS) to assist one another by recognizing and responding to potential opportunities for each other’s goals [2]. In theory, multi-agent systems can clearly benefit from the ability of agents to act opportunistically. In practice, however, taking advantage of an opportunity at an inter-agent level is far from trivial: The agents should be capable of recognizing whether a given event or situation may be an opportunity for a goal of another agent in the system, and of responding appropriately to these recognized opportunities. Two key issues can make the potential practical attractiveness of multi-agent opportunism somewhat questionable. First, both recognizing opportunities and responding to them should have low computational complexity, otherwise the MAS will be more “socially friendly” than useful. Second, in order for an agent to recognize potential opportunities for other agents, the agent clearly has to know something about what these other agents are doing. We believe, however, that in many applications this “something” will tend to be very limited. The question is whether multi-agent opportunism can be effective in such cases of extremely limited shared knowledge?

The Radsim Simulator

In this chapter we describe Radsim (Radar simulation), a simulation environment developed to support the ANTs common challenge problem. Topics discussed include the general simulation model used, the models of the Doppler sensors and moving targets, the communication model used among the agents, the control of the system through an external API, and the support for conducting experiments.

Asynchronous chess competition

Asynchronous Chess (AChess) is a platform for the development and evaluation of real-time adversarial agent technologies. It is a two-player game using the basic rules of chess with the modification that agents may move as many pieces as they want at any time. Modifying chess in this way creates a new robust, asynchronous, real-time game in which agents must carefully balance their time between reasoning and acting in order to out-perform their opponent. As a fast-paced adversarial game, many challenges relevant to real-world applications arise which give it merit for study and use.

Agent-Supported Planning in Distributed Command and Control Environments

To be able to meet the future challenge of employing forces anywhere in the world in support of national security objectives, modern military forces require highly synchronized, distributed planning and re-planning capabilities that are sufficiently flexible to adapt to any level of conflict. This talk will present a research program underway at the USAF Research Laboratorys Information Directorate known as DEEP (Distributed Episodic Exploratory Planning). DEEP is an agent-based distributed planning system that has been designed to support future military command and control (C2) operations. The talk will discuss the motivation for moving from a centralized planning model to a distributed mixed-initiative approach, along with the DEEP architecture and the key research challenges for achieving this vision. The distributed agent-supported planning capabilities, which utilize past experience to solve current problems, will be emphasized.