When an engineer needs to 'think outside the box', it is often useful for him or her to examine how nature addresses a similar design problem. Reliable separation of aircraft is a problem very similar to that solved by flocking and schooling animals, and is related to that solved by bats when capturing flying insects. In all these cases, sensory data is used in a fairly direct manner to control the position and velocity of a physical body. Natural systems differ from current air traffic control systems in assigning the separation management role to the individual animal instead of to a third party, but the evolution of air traffic control towards free flight now proposes to place more of this responsibility on the pilot, suggesting that an understanding of how animals maintain or eliminate separation is useful.

The natural solution to this problem works well. The Mexican free-tailed bat (Tadarida brasiliensis) lives in colonies with millions of bats in some warm caves of the southwest United States, emerging twice each night to hunt flying insects. Hundreds of bats per second make up the traffic stream passing through the cave entrance; yet there is no accumulation of dead and wounded bats on the ground below. Possible explanations for this include rapid reaction times and high maneuverability to avoid conflicts at the last second; yet the reaction time of bats is on the order of 100-300 milliseconds—not a great deal better than that of humans—their maximum horizontal acceleration in the laboratory appears to be on the order of 1.5 Gs, and the active echolocation range for bat-sized targets is only a few meters.

So how do they do it? This paper examines what is currently known about how bats and other animals control their flight to maintain separation or capture targets.

The Mexican or Brazilian free-tailed bat is a medium-sized (10-15 gm) brown bat with broad ears, large feet, and a tail half free of the flight membrane that connects the legs. In the United States, the eastern subspecies is non-migratory, but the western subspecies migrates yearly between Mexico and the southwest. This subspecies roosts in caves, mine tunnels, old wells, hollow trees, houses, bridges, and other buildings in groups of several dozen to several million, forming great clusters that blanket huge areas of the walls and ceilings of their retreat. Tadarida brasiliensis is a fast-flying bat, being observed at over 80 miles per hour (110 feet per second) when gliding back into Carlsbad Caverns (Barbour & Davis, 1969), and has long, narrow wings with a span of about a foot that are well-adapted to flight in open areas (Norberg, 1987). They hunt at night, leaving the roost in a swarm that often forms a single column 30 feet in diameter. They accelerate gradually, reaching 60 miles per hour (88 feet per second) at the point the swarm breaks up (Barbour & Davis, 1969; Schmidly, 1991). The appearance of these columns has led to the term 'smoke hole' being applied to the caves where they roost.

At Carlsbad Caverns, the free-tailed bat population was estimated at 352,000 during the fall of 1996 using infrared photographs of the ceiling of Bat Cave (National Park Service, 2000). This is a decrease from the 8.7 million estimated by Vemon Allison (1937), who observed a flight column with a diameter of approximately 20 feet, and a time length of about 16 minutes. Allison made his estimate assuming a density of one bat per cubic foot in the column, and a speed of about 29 feet per second. David Roemer (1999) reduced this estimate to 3.6 million bats by assuming the observed flight speed to be about 12 feet per second. The estimated separation of about a foot increases to 2.2 feet if the population estimate is reduced to 360,000 bats, but in either case, bats in the swarm are spaced at about wingtip distance.

It is unlikely that the bats in the swarm use echolocation or sonar for localization, since the cries of several hundred nearby bats are likely to swamp the echoes that the individual bat is listening for. Bats have well-developed eyes, comparable in sensitivity to those of other mammals (Nowak, 1994), and—although believed to play a supplementary role in the daily lives of these bats—vision is probably the primary sense used to maintain separation while flying in the swarm.

Bats do not behave like small homing missiles, but control their flight very much like human pilots. Their motor reaction time is on the order of 0.1-0.2 sec (Casseday & Covey, 1996), about twice as fast as that for humans, and they seem to plan their maneuvers about 0.8-1.0 sec in advance (Erwin, Wilson, & Moss, 2000), which is comparable to the emergency response time for humans. A human pilot, given comparable aerodynamic performance, should be able to come close to matching the performance of a bat.

Tadarida brasiliensis does poorly in captivity, but flight room studies of the house bat, Eptesicus fuscus, a slower but comparably sized bat, do provide some insight into the aerodynamic performance of bats (Erwm et al., 2000). Despite the general impression that bats are acrobatic, Eptesicus fuscus limits the G force it pulls to about 1.5 Gs in the flight room. It also limits its acceleration and deceleration to about 0.4 Gs, using vectored forward thrust instead of drag to slow. Flapping flight aerodynamics allow it to generate sufficient lift to hover for short periods. Probably the most similar aerodynamic environment for a human pilot would be a helicopter or a Harrier jump jet (AV8B), but the aerodynamic constraints faced by the bat, especially while flying in the swarm, are similar to those faced by most human pilots. Note that the bat's motor reaction time of 0.1-0.2 sec is probably enough to handle gradual separation changes in the swarm, but the 0.8-1.0 sec planning time is too big for handle emergencies.

So in the swarm, bats probably avoid collisions by flying in formation, avoiding sudden maneuvers, and maintaining constant situational awareness. Given the large amount of sensory data to be processed, they need some mechanism for selecting only what is important in this task. This selection can be based on habituation or by predicting and ignoring safe nearby behavior.

There are some data that provide some insight into how bats perceive their world. These were first reported by Mohres and Ottingen-Spielberg (1949) using the following terms:

• • Erstorientierung—the orienting reaction when bats first encounter a novel situation.
• • Wiederorientierung—how bats behave when they fly in a familiar space.

These phenomena were observed in the behavior of a bat that was accustomed to roosting in a cage in a room. When the cage door was opened, it flew around the room for a short time and then returned to its perch. While the bat was flying, the researchers rotated the cage or removed it, and noted that the bat continued to behave as if the cage were in its normal position. This suggested that bats use and maintain a world model that is only modified to match reality if circumstances force it to reorient-

Rawson and Griffin investigated this further (Griffin, 1958, 1988), asking whether bats even needed to make their cries at all. This experiment involved placing and moving obstacles in a flight room. They found that the bat still cried, but seemed to ignore the echo returns, habituating to the original environment to a degree that large scale changes were required to trigger the orienting reaction.

The implication of this work was that bats seemed to have an internal world model - a predictive planning process—that was updated based on sensory data only when significant novelty was detected. There appears to be a match/mismatch process (Pribram, 1971) that filters the large amount of data that the bat's echolocation cries generate down to unexpected events that are likely to matter to the bat. A similar pattern of habituation to the environment is seen in most vertebrates; bats only take it further.

Research by Curtis Bell and his colleagues have provided some insight into the neural mechanisms that may underlie this match/mismatch process (Bell, Bodznick, Montgomery, & Bastian, 1997; Bell, Caputi, & Grant, 1997; Bell, Han, Sugawara, & Grant, 1997). They had studied the electros ensory system of mormyrid fish, controlling the sensory input very precisely, and they demonstrated that dynamic sensory expectations derived from past sensory input were observable and were being subtracted from current sensory input, allowing novelty to stand out more clearly (Bell, 2000). Their results were similar in four distinct groups of fishes and so are thought to generalize to cerebellum-like structures in other sensory systems (such as the early auditory system) and taxa.

The difficulty with applying Bell's result to the auditory system of bats is that electric fish take several minutes to memorize and habituate to the local environment, and they have to remain stationary to detect novelty. Bats, on the other hand, fly at 10-12 feet per second in the flight room (and faster in the field), so that they can't simply memorize the pattern of acoustic stimuli associated with a position in me room. Instead, they have to predict the pattern of stimuli based on where they were when the made their cry and where they are when they hear the echo. This requires a fast modeling process to predict the future acoustic environment rather than a memory process to compare the immediate past with the present.

The modeling reported in Erwin, Wilson, and Moss (2000) showed that a match/mismatch process similar to that seen in electric fish was likely to play a role in auditory localization. In this study, a computational sensorimotor model of target capture behavior by an echolocating bat was developed. It integrated acoustics, target localization processes, flight aerodynamics, and target capture planning to produce model trajectories similar to those observed in behavioral trials. Estimates of range were based on echo delay, azimuth on the relative intensity of the target in the two ears, and elevation on the spectral pattern of the echolocation return in a match/mismatch process. Flapping flight aerodynamics were used to produce realistic model trajectories. Nonpredictive tracking of the target, using only azimuth and elevation to control the intercept, was shown to be inadequate. Target capture using maneuvering flight was generally successful when the model's path was controlled by a planning process making use of a predictive simulation.

These results provide some insight into how to engineer air traffic control. Even at high aircraft densities, the pilot has enough time to react to the developing situation as long as traffic can be relied upon to fly in approximate formation and the response to emergency situations is coordinated so that evasive actions do not create additional emergencies.

The author wishes to thank Annemarie Suriykke, Cynthia F. Moss, Bernard Halprin, and Wilson Felder for discussion and comments. This preparation of this paper was funded by FAA Contract DTFAO 1 -95 -C-00031.

Allison, V. C. (1937). Evening bat flight from Carlsbad Caverns. Journal of Mammalogy, 18, 80-82. Barbour, R. W., & Davis, W. H. (1969). Bats of America. Lexington, Kentucky: The University Press of Kentucky.

Bell, C. C. (2000). OGHS Faculty Web Pages. Bell, C. C., Bodznick, D., Montgomery, J., & Bastian, J. (1997). The generation and subtraction of sensory expectations within cerebellum-like structures. Brain,Behavior and Evolution, 50(Suppl 1), 17-31.

Bell, C. C., Caputi, A., & Grant, K. (1997). Physiology and plasticity of morphologically identified cells in the mormyrid electrosensory lobe. Journal of Neuroscience, 7 7(16), 6409-6423.

Bell, C. C., Han, V. Z., Sugawara, Y., & Grant, K. (1997). Synaptic plasticity in a cerebellum-like structure depends on temporal order. Nature, 387,278-281.

Casseday, J. H., & Covey, E. (1996). A neuroethological theory of the operation of the inferior colliculus. Brain, Behavior and Evolution, 47, 311-336.

Erwin, H. R., Wilson, W. W., & Moss, C. F. (2000). A computational sensorimotor model of bat echolocation. Journal of the Acoustical Society of America. Griffin, D. R. (1958). Listening in the Dark. Ithaca, New York: Comstock Publishing Associates.

Griffin, D. R. (1988). Cognitive aspects of echolocation. In P. E. Nachtigall & P. W. Moore (Eds.), Animal Sonar: Processes and Performance (Vol. 156, pp. 683- 690). New York: Plenum Press.

Mohres, F. P., & Ottingen-Spielberg, T. (1949). Versuche liber die Nahorientierung und das Heimfindvermogen der Fledermause, Verhandlungen der deutschen Zoologen in Mainz (Vol. 43, pp. 248-252).

National Park Service. (2000). Carlsbad Caverns National Park: Natural Resources: Bats: Counting the Bats.

Norberg, U. M. (1987). Wing form and flight mode in bats. In M. B. Fenton & P. Racey & J. M. V. Rayner (Eds.), Recent advances in the study of bats (pp. 43-56). Nowak, R. M. (1994). Walker's Bats of the World. Baltimore, Maryland: The Johns

Hopkins University Press. Pribram, K. H. (1971). languages of the brain: experimental paradoxes and principles in neuropsychology. New York: Brandon House, Inc.

Roemer, D. (1999). Counting bats at Carlsbad Cavern: the 8.7 million number. Canyons and Caves, pp. 1. Schmidly, D. J. (1991). The bats of Texas ( 1st ed. Vol. 11). College Station, Texas: Texas A&M University Press.