1. Introduction

Two trends in electronic hardware are expected to have significant impacts on day-to-day operational practice in the military and in most other organizations that operate large numbers of resources: (i) more 'intelligence' is being embedded into smaller and cheaper devices; (ii) wireless communication is improving in availability, performance and cost. As a result of these trends, individual, small devices are becoming more capable, but even more so, collaborative networks of small devices are beginning to rival large, traditional resources in terms of capability (at least in some dimensions) and to exceed them in terms of cost-effectiveness and flexibility.

This introduction will briefly discuss a couple of motivating examples and then address one of the main problems associated with large networks of resources: how to manage the resources in a scalable, robust and cost-effective manner.

1.1 Examples

1.1.1 Sensor Networks

Traditionally, sensor networks have been comprised of a small number of high-quality, high-cost sensors connected to a central observation and control station using either wired communication or high-power radio. Examples include observational satellites, closed-circuit television and radar installations. While such networks continue to be valuable, the increasing availability of lower-quality, lower-cost sensors and short-range radio communication provides an alternative in which many "small" sensors collaborating replace a single, "large" sensor.

What advantages might be gained by using many smaller sensors?

Local Coverage
Because many sensors can be deployed, it may be possible to achieve coverage of regions that would be obscured from the field of view of traditional sensors; for example, inside buildings or under foliage.
Robustness
If properly designed and managed, a network of many small sensors does not have critical failure points: disrupting the network would require disabling a large fraction of the sensors.
Flexibility
If the sensors are mobile, some of them can be repositioned to closely examine regions of interest as they arise, while the remainder continue with broad-area surveillance.
Cost savings
For some applications, a network of many small sensors may be much cheaper than an equivalent network of large sensors.

There are, of course, several limitations to using small sensors and several potential problems that need to be addressed.

Broad-Area Coverage
For some applications, large sensors such as satellites can surveil areas that are much too large to be covered by a network of small sensors like those currently being developed. Of course, future developments may produce small sensors that have much larger areas of coverage, allowing equivalent resolution to be achieved with a reasonable number of sensors. Nevertheless, it seems reasonable to consider networks of small sensors as complementary to networks of large sensors.
Energy Supplies
Currently available small sensors are battery-powered. Current research is investigating the use of solar power to recharge batteries in the field. Until such technology becomes feasible and cost-effective, an important problem is how to manage the sensors so as to reduce their energy consumption and thus extend the time they can be deployed without maintenance.
Data Fusion
Because the data gathered by each small sensor is typically relatively low in quality, typical applications require data from several or many sensors be fused to provide sufficiently accurate information. Data fusion for sensor networks is an active research area.
Sensor Coordination
Some sensors are multi-modal: each such sensor is capable of measuring several different physical phenomena (sound, light, magnetism, vibration, for example) or is capable of focusing its measurements (along different directions or frequencies, for example). Moreover, some sensors are mobile and the data they gather is strongly dependent on position. Consequently, the sensors must be coordinated so that their various modes either complement each other (to achieve coverage) or reinforce each other (to achieve high-quality data after fusion).

This last area, coordination, is the focus of the ANTs program: given a sensor network, coordinate the sensors to accurately detect and monitor features that are of interest. For example, an "feature" may be a moving object and the network may be required to detect and track objects as they move into and through its surveillance region.

1.1.2 Drone Teams

Just as new generations of small sensors may be used en masse instead of traditional, large sensors, so too new generations of small actuators may be used instead of traditional, large actuators. For example, unmanned air vehicles (UAV) and, in particular, the combat variants (UCAVs) are much smaller and cheaper than manned aircraft but are capable of undertaking some of the same tasks, to a limited extent (their payloads are much smaller, for example).

Because of the limitations of a single UAV, it is useful to have UAVs operate collaboratively. Over open terrain, UAV teams may operate as small, tightly-knit squadrons with narrow goals. However, in urban environments, it may be more useful to have many UAVs operate as a loosely organized network, opportunistically collaborating as the need arises to accomplish (possibly competing) goals that are dynamically generated by many ground commanders, to surveil broad regions for hostile activity and to maintain communication infrastructure.

Under such operating circumstances, many of the same pros and cons apply as for sensor networks, although generally energy consumption tends to be much higher and a single UAV would be deployed for only a limited time before maintenance.

1.2 Problem Statement

To summarize, the problem that is the main focus of the ANTs program may be described as follows:

In addition, the following requirements hold:

Even ignoring the communication aspects, coordinating a large number of resources is a difficult problem due to the combinatorial growth in the number of possible states of the network. Given that the ANTs program requires real-time adaptivity, it is clearly infeasible to try to attain perfect coordination. Instead, the aim is to achieve coordination that is good enough for the resource network to accomplish most of its tasks without being too inefficient, and to produce the coordination decisions quickly enough that the resource network can keep pace with a dynamic environment.

This objective was summarized in the slogan: good-enough, soon-enough.

The ANTs program also proposed a conceptual architecture for coordination, that was, in the main, respected by the various projects. Because the size of the network and the communication latency make it impossible for any one process in the network to acquire global knowledge, the coordination mechanism would be split into many processes, each operating with limited knowledge and authority.

Moreover, because the resource network itself may be dynamic (as the status of resources changes) and because the tasks that the network is to accomplish may be dynamic, the organization of the coordination processes would also be dynamic: in effect, the processes would dynamically form themselves into teams to collectively accomplish tasks.

In order for the processes to efficiently form effective teams, they would need to communicate with each other. Since, each process would likely have somewhat different information and responsibilities from the others, due to its particular location in the network, a real-time negotiation framework seemed well-suited, as it would provide mechanisms for reaching a compromise with limited effort in spite of non-uniform and possibly competing constraints and objectives.

Thus the program name: Autonomous, Negotiating Teams.

Program Organization

The program was organized as follows:

Two main challenge problems
To provide demonstration and evaluation test beds, two challenge problems were identified: (i) a logistics challenge problem based on management of resources in a flight wing; (ii) an electronic warfare (EW) challenge problem based on managing networked sensors.
Logistics projects
These projects investigated problems in and solutions to the logistics challenge problem.
EW projects
These projects investigated problems in and solutions to the EW challenge problem.
Dynamics and complexity projects
The projects focused on general properties of the classes of problems and solutions arising from the logistics and EW projects. For example, constraint satisfaction problems model many aspects of resource coordination and show interesting phase transitions for large problems, whereby the solutions space shows discontinuities as certain parameters of the problem are varied.