Insect Behaviour: Controlling Flight Altitude with Optic Flow
2
Dispatches Insect Behaviour: Controlling Flight Altitude with Optic Flow Insects can smoothly control their height while flying by adjusting lift to maintain a set-point in the ventral optic flow. The efficacy of this simple flight-control mechanism has been demonstrated using a robot helicopter. Barbara Webb When trying to understand animal behaviour it is common to think about how information is extracted from the environment and an internal representation of that information used to calculate an appropriate controlled response. But this approach can be misleading when we assume that certain information is required to perform a task, when in fact a simpler solution, exploiting the closed-loop interaction of the animal with its environment, is available. Wehner [1] has described such solutions, where the cue required to perform the task is directly picked up by the perceptual system, as ‘matched filters’. An example is discussed in the article in this issue by Franceschini et al. [2]. The basic problem addressed by Franceschini et al. [2] is how a flying insect controls its height above the ground, for example in take-off, landing and maintaining a constant height above varying terrain. Does the insect measure its altitude, and if so, how? It could potentially extract height information from the ventral optic flow: basically, the higher it is, the slower will be the apparent motion of the ground directly below it. But to actually extract height information from this signal, the insect would need to know its true velocity relative to the ground — its groundspeed — as the ventral optic flow is proportional to its groundspeed divided by its height. In their paper, Franceschini et al. [2] describe a flight control solution in which neither groundspeed nor altitude are explicitly determined. Instead, the directly available cue — ventral optic flow — is used in a feedback control loop, with the insect altering its lift to maintain a setpoint ventral optic flow, and thus a constant groundspeed:height ratio. This simple mechanism has a number of desirable properties. If the animal increases its forward speed, it will automatically increase its height as it takes off. If it gradually decreases speed, it will gradually decrease height and thus land smoothly. If the terrain rises, the ventral optic flow will increase and the insect will compensate by increasing its height. If the insect is slowed by a headwind, it will descend; this is a strategy likely to reduce the headwind, or even to lead the insect to land if it cannot make any progress against the headwind. Franceschini et al. [2] describe reports of each of these effects having been observed in insect flight. Most of these reports are qualitative — such as the observation that bees flying over mirror-smooth water, which provides no ventral optic flow, may descend so far that they end up in the water and drown [3]. In the case of landing, there are quantitative data [4] which closely fit the predictions of their model. While such a scheme sounds intuitively plausible, Franceschini et al. [2] have also shown that it can work in practice by implementing it on a robotic helicopter. This uses just two photoreceptors, coupled to produce an elementary motion detector based on fly vision. Forward speed is altered by changing the pitch of the helicopter rotor, and lift by changing the rotor speed. This physical model has been designed to match characteristics of insects such as the angle between photoreceptors, the flight-speed range and optical flow sensor range. It thus shows that the controller is plausible within the real physical constraints known to hold for insects. This approach — a biorobotic evaluation of a biological hypothesis — is becoming increasingly popular. Recent examples include robotic investigations of locomotion in animals as diverse as water striders [5], salamanders [6] and humans [7]. It allows modellers to close the loop: that is, to understand how actions affect the interaction with the world that leads to sensory input; and to understand this as a dynamic loop rather than as cause-and-effect. It can also close the loop from biology to engineering by producing new predictions and insights that can be used in biology [8]. Indeed, the idea of using optic flow directly to control behaviour has been explored in a number of other insect-inspired robots. Lateral optic flow can be used to regulate forward speed [9], possibly in a similar feedback loop to the lift control described above. Because lateral optic flow depends on the distance of the surrounding objects, such a controller will adaptively slow the animal when it is traversing narrower apertures or slow a robot moving through cluttered environments [10]. Balancing the lateral optic flow on each side also results in centring [10]. And expansion in the optic flow field is an indication of impending collision: by triggering a response when the rate of expansion increase is above some threshold, collisions can be avoided, again without any explicit detection of the actual size or distance of the object. Specialised neurons to detect this signal have been described in Current Biology Vol 17 No 4 R124
-
Upload
barbara-webb -
Category
Documents
-
view
212 -
download
0
Transcript of Insect Behaviour: Controlling Flight Altitude with Optic Flow
Current Biology Vol 17 No 4R124
Dispatches
Insect Behaviour: Controlling Flight Altitude withOptic Flow
Insects can smoothly control their height while flying by adjusting lift tomaintain a set-point in the ventral optic flow. The efficacy of this simpleflight-control mechanism has been demonstrated using a robothelicopter.
Barbara Webb
When trying to understand animalbehaviour it is common to thinkabout how information is extractedfrom the environment and aninternal representation of thatinformation used to calculate anappropriate controlled response.But this approach can bemisleading when we assume thatcertain information is required toperform a task, when in facta simpler solution, exploiting theclosed-loop interaction of theanimal with its environment,is available. Wehner [1] hasdescribed such solutions, wherethe cue required to perform thetask is directly picked up by theperceptual system, as ‘matchedfilters’. An example is discussed inthe article in this issue byFranceschini et al. [2].
The basic problem addressed byFranceschini et al. [2] is how a flyinginsect controls its height above theground, for example in take-off,landing and maintaining a constantheight above varying terrain. Doesthe insect measure its altitude, andif so, how? It could potentiallyextract height information from theventral optic flow: basically, thehigher it is, the slower will be theapparent motion of the grounddirectly below it. But to actuallyextract height information from thissignal, the insect would need toknow its true velocity relative to theground — its groundspeed — asthe ventral optic flow isproportional to its groundspeeddivided by its height.
In their paper, Franceschini et al.[2] describe a flight control solutionin which neither groundspeed noraltitude are explicitly determined.Instead, the directly available
cue — ventral optic flow — isused in a feedback control loop,with the insect altering its lift tomaintain a setpoint ventral opticflow, and thus a constantgroundspeed:height ratio. Thissimple mechanism has a number ofdesirable properties. If the animalincreases its forward speed, it willautomatically increase its height asit takes off. If it gradually decreasesspeed, it will gradually decreaseheight and thus land smoothly. Ifthe terrain rises, the ventral opticflow will increase and the insect willcompensate by increasing itsheight. If the insect is slowed bya headwind, it will descend; this isa strategy likely to reduce theheadwind, or even to lead theinsect to land if it cannot make anyprogress against the headwind.Franceschini et al. [2] describereports of each of these effectshaving been observed in insectflight. Most of these reports arequalitative — such as theobservation that bees flying overmirror-smooth water, whichprovides no ventral optic flow, maydescend so far that they end up inthe water and drown [3]. In the caseof landing, there are quantitativedata [4] which closely fit thepredictions of their model.
While such a scheme soundsintuitively plausible, Franceschiniet al. [2] have also shown that it canwork in practice by implementing iton a robotic helicopter. This usesjust two photoreceptors, coupledto produce an elementary motiondetector based on fly vision.Forward speed is altered bychanging the pitch of the helicopterrotor, and lift by changing therotor speed. This physical modelhas been designed to matchcharacteristics of insects such as
the angle between photoreceptors,the flight-speed range and opticalflow sensor range. It thus showsthat the controller is plausiblewithin the real physical constraintsknown to hold for insects.This approach — a bioroboticevaluation of a biologicalhypothesis — is becomingincreasingly popular. Recentexamples include roboticinvestigations of locomotion inanimals as diverse as waterstriders [5], salamanders [6] andhumans [7]. It allows modellers toclose the loop: that is, tounderstand how actions affect theinteraction with the world thatleads to sensory input; and tounderstand this as a dynamic looprather than as cause-and-effect. Itcan also close the loop frombiology to engineering byproducing new predictions andinsights that can be used inbiology [8].
Indeed, the idea of using opticflow directly to control behaviourhas been explored in a number ofother insect-inspired robots.Lateral optic flow can be used toregulate forward speed [9],possibly in a similar feedback loopto the lift control described above.Because lateral optic flow dependson the distance of the surroundingobjects, such a controller willadaptively slow the animal when itis traversing narrower apertures orslow a robot moving throughcluttered environments [10].Balancing the lateral optic flow oneach side also results in centring[10]. And expansion in the opticflow field is an indication ofimpending collision: by triggeringa response when the rate ofexpansion increase is above somethreshold, collisions can beavoided, again without anyexplicit detection of the actual sizeor distance of the object.Specialised neurons to detectthis signal have been described in
DispatchR125
the locust and modelled on a robot[11]. The same principle has beenused to enable a small free modelaircraft to avoid collisions withwalls [12].
Exploiting some aspect of opticflow directly to control locomotionis often discussed, followingGibson [13], as an example of an‘affordance’. The exact meaning ofthis term is subject to debate, butessentially it is the idea that whatanimals are designed to perceiveare opportunities for action, ratherthan action-neutral properties ofthe environment around them.Instead of seeing shape, size anddistance of an object, for example,we observe its graspability. Thisinfluential, and sometimescontroversial, view of perception isparticularly relevant to robotics,where specialised sensorysystems for cues such as opticflow have often proved more usefulthan conventional computervision. In one sense these may bethought of as tricks or short-cutsthat enable the animal or robot toavoid difficult measurementsand calculations. But the conceptof action-oriented perceptionmay also be important inunderstanding higher levelcognitive skills, as it stronglydetermines how we structure theworld around us.
For insect flight control, manyissues remain to be resolved.Ventral optic flow can be easilydetected if the animal is flyingstraight ahead and the sensor ispointing straight down. But if the
Speciation GenetiConflict and the O
Evolutionary biologists have long recinviability of species hybrids must invinteractions between two (or more) ghybrid incompatibility genes has now
Daven C. Presgraves
Species are often reproductivelyisolated from one another by theintrinsic sterility or inviability oftheir hybrids. Darwin devoted anentire chapter of his Origin of
insect pitches, rolls or rotates,ventral optic flow will be distorted.Can the animal measure anddiscount these movements, or areother sensorimotor loops, such asthe optomotor reflex, deployedsimultaneously to minimise them[14]? Is there any evidence ofsystematic difference, for examplein the sensitivity range, ofelementary motion detectorspointing at different parts of thevisual field [15] that would fit withthe proposed difference in controlfunction? What exactly are thewing movements that need to becontrolled [16] and might thesealso be ‘matched’ to specificcontrol problems? Willunderstanding the basic controlrules help us to trace out the neuralpathways that support thisbehaviour? The combination ofbehavioural experiments androbot models is likely to be animportant tool in futurediscoveries.
References1. Wehner, R. (1987). ‘Matched filters’ —
neural models of the external world.J. Comp. Physiol. A. 161, 511–531.
2. Franceschini, N., Ruffier, F., and Serres, J.(2007). A bio-inspired flying robot shedslight on insect piloting abilities. Curr. Biol.17, 329–335.
3. Heran, P., and Lindauer, M. (1963).Windkompensation undSeitenwindkorrektur der Bienen beimFlug }Uber Wasser. Z. vergl. Physiol. 47,39–55.
4. Srinivasan, M.V., Zhang, S., Chahl, J.S.,Barth, E., and Venkatesh, S. (2000).How honeybees make grazing landingson flat surfaces. Biol. Cyb. 83, 171–183.
5. Hu, D.L., Chan, B., and Bush, J.W.M.(2003). The hydrodynamics of waterstrider locomotion. Nature 424,663–666.
cs: Epistasis,rigin of Species
ognized that the sterility andolve incompatible epistaticenes. The first pair of suchbeen identified.
Species [1] to hybrid sterilitybecause, being scrupulous, hewished to confront the possibleshortcomings of his theory headon: why would natural selection,which acts to increase individualfitness, cause the evolution of
6. Ijspeert, A.J., Crespi, A., andCabelguen, J.M. (2005). Simulation androbotics studies of salamanderlocomotion. Applying neurobiologicalprinciples to the control of locomotionin robots. Neuroinformatics 3,171–196.
7. Collins, S., Ruina, A., Tedrake, R., andWisse, M. (2005). Efficient bipedal robotsbased on passive dynamic walkers.Science 307, 1082–1085.
8. Webb, B. (2000). What does roboticsoffer animal behaviour? Anim. Behav. 60,545–558.
9. Baird, E., Srinivasan, M., Zhang, S., andCowling, A. (2005). Visual control of flightspeed in honeybees. J. Exp. Biol. 208,3895–3905.
10. Srinivasan, M.V., Chahl, J.S., Weber, K.,Nagle, M.G., and Zhang, S.W. (1999).Robot navigation inspired by principles ofinsect systems. Robotics Auton. Sys. 26,203–216.
11. Blanchard, M., Rind, F.C., andVerschure, P.F.M.J. (2000). Collisionavoidance using a model of the locustLGMD neuron. Robotics Auton. Sys. 30,17–38.
12. Zufferey, J.C., and Floreano, D. (2006).Fly-inspired visual steering of an ultralightindoor aircraft. IEEE Trans. Robotics 22,137–146.
13. Gibson, J.J. (1966). The sensesconsidered as perceptual systems(Boston: Houghton Mifflin).
14. Neumann, T.R. and Bulthoff, H.H., (2001).Insect inspired visual control oftranslatory flight. Proceedings ofAdvances in Artificial Life: 6th EuropeanConference, ECAL 2001, Prague, CzechRepublic, September 10–14.
15. Egelhaaf, M., Kern, R., and Krapp, H.G.(2002). Neural encoding ofbehaviourally relevant visual-motioninformation in the fly. Trends Neurosci.25, 96–102.
16. Dickinson, M.H. (2006). Insect flight. Curr.Biol. 16, 309–314.
Institute of Perception, Action andBehaviour, University of Edinburgh,JCMB, Kings Buildings, Mayfield Road,Edinburgh, Eh9 3JZ, UK.E-mail: [email protected]
DOI: 10.1016/j.cub.2006.12.008
hybrid sterility? As with manyproblems, Darwin struggled withthe genetic details but, in the end,got the basics right: hybrid sterility‘‘is not a specially endowedquality, but is incidental on otheracquired differences,’’ (p. 245)and is caused by a hybrid’s‘‘organization having beendisturbed by two organizationshaving been compounded intoone’’ (p. 266).
Fifty years would pass beforeBateson [2] and later Dobzhansky[3] and Muller [4,5] deviseda genetic model for the evolutionof such hybrid fitness problems
