Swarm Intelligence

By Gerard P. Gilfoyle

Swarm intelligence is a potentially fertile ground for new methods for countering terrorism. A group of non-intelligent agents—robots, sensors, etc.—interact with their environment and each other to produce collectively intelligent behavior. An ant colony is the prototypical example from nature. The individual agents (ants) exhibit little innate intelligence and often choose the "wrong" solution, such as the long path to a food source. Nevertheless, using independent communication among themselves via pheromone trails, the majority of the colony finds the best solution, i.e., the shortest path to a food source. The swarm intelligence emerges from the multiple interactions of the agents and enable it to discriminate among many paths toward a goal.

Several characteristics distinguish swarm intelligence from other systems. The individual components or agents are distributed and autonomous; there is no central, guiding hand determining the actions of each agent. Each agent carries sensors or actuators to perceive and/or change its environment and communicate. The agents have no explicit model of their environment; there is no detailed "roadmap" telling them where to go. The decision-making of the group is distributed across many platforms and devices while the systems environment guides the overall actions of the swarm.

These ideas are already being applied. A robot search team (called the "Snowbots") developed at Sandia National Laboratory can locate avalanche victims in snow four times faster than humans using dogs. Telecommunications firms are using computational "ants" (agents) to produce faster, more robust communications networks. Transportation firms use these algorithms to pick the best way to rout gasoline trucks.

These techniques could apply to many phases of the war on terrorism. Finding nuclear material hidden in a city is still a daunting task. Radiation detectors are hindered by widely changing background radiation from natural sources that can produce many false positives. If the alarms go off too often, the users will just turn them off. An array of sensors distributed around sensitive areas, such as the Mall in Washington, DC, could detect nuclear radiation leaking from a radiological dispersal device or "dirty bomb" before it explodes. This array can cover a large area and correlations in time and space among multiple "hits" in different detectors provide a powerful filter to reduce false alarms. The Defense Science Board, which advises the U.S. Department of Defense, has encouraged the development of such a capability. Mitigating the effects of an attack is essential for countering terrorism. If an attack has just occurred, then a swarm of mobile robots could enter hazardous environments, like a collapsed building or a contaminated area. They could assess damage, locate survivors, and search for radiological, chemical or biological agents.

The enabling technologies of swarm intelligence cut across many disciplines and include areas relevant to the members of the APS. The component technologies, in particular, used in the agents (the robots or sensors) are often still in their infancy. What is impressive now is that many of the concepts have been demonstrated with rather unsophisticated agents. Improvements in sensors or actuators would drive the technology forward. A partial list of the relevant technologies include:

  • Miniature and micro-robotic devices
  • Micro-electro-mechanical devices (MEMS)
  • Sensor technology for chemical and biological agents
  • Micro manipulation devices
  • Low-power, long-range communications
  • Enhanced power supplies
  • Mobility approaches: hopping versus gliders, etc.
  • Fabrication of miniature, micro- and nanoscale devices at low cost

The control technologies (software) used to guide the swarm are ripe for further development. Each agent must be programmed to act independently, sense and respond to its surroundings, communicate the correct information to the right machine or person, or even take some action. How well can this be done with only limited guidance from a central command? How is a deluge of data relayed from the agents in the field interpreted? How and when does this system decide to alert humans to a problem?

There is a need to develop and apply mathematical tools to analyze and predict the behavior of a swarm. Arrays of robots will only be used if they can be trusted to act as expected and not in some unanticipated, harmful way. What happens to the swarm when it loses a significant number of agents? Does the performance decline "gracefully" or does it change precipitously? Ants die regularly in a colony, but the colony continues to thrive. The problem here may be to understand the dynamics of a system of many particles under equilibrium and non-equilibrium conditions. One can speculate that expertise in many-body physics and complexity theory could apply.

Many agents will likely use radio frequency communications, but our understanding of signal propagation loss is incomplete. An array of sensors tossed from a helicopter will end up in different orientations on the ground. If a sensor's antenna is lying on the ground, what is the maximum range of the radio in that terrain, and thus how dense should the sensor network be? What is the signal loss near the ground, in hilly country versus flat land, or even in collapsed buildings?

Hordes of mobile, interacting agents may have to maintain a communications network among themselves and with humans. Consider robots inspecting a cargo ship for nuclear material that must cooperate to find a small radiation signal from a large changing background. Can we produce a robust network when the "nodes" are moving over rugged, unknown terrain? Imagine a cell-phone network with the base stations carried by hikers wandering the countryside.