PS3060: Perception and Action
Term II, MONDAY 10–12 am (Room 128 Wolfson)

Lecture 2: A view of the world from the cockpit of a fly – the ecology of navigation

Course co-ordinator: Johannes M. Zanker, j.zanker@rhul.ac.uk, (Room 218)


Topics Lecture 2:


The ecological approach to vision

Gibson (1950, 1986) drew attention on two aspects of vision, which gave rise to a novel approach to understand perception

 a carpet of clouds seen from above

what are the direct distance cues ?

recent developments in this area extend the scope : 'visual ecology'
looking for specific adaptations of sensory and motor systems to particular environment & lifestyle, interfacing with comparative neuroscience


The optic array

central to Gibson's theory: the distribution of light (intensity, wavelength, etc.) from a particular viewpoint (= optic array) contains all information available; it is a rich and immediate source, because it informs about the surroundings and the relative position of the observer to the surroundings




sketch of light distribution in a simple environment, as seen from two particular viewpoints


effects of changing location: characteristic shifts / changes of the light distribution in the optic array: displacements, occlusions, shape intensity texture changes


Flowfields

after almost 50 years perhaps the  most important aspect of Gibson's work

continuous change of position (observer movement) generates characteristic patterns of image motion, that should directly indicate observer motion and thus an be used for control of locomotion: flowfields

(this is a specific problem from Gibson's professional context: training and selection of able pilots : fast responses require 'direct perception')

ideal picture perhaps more obvious in other species than human:
more low-level processing, natural adaptation to flying, more generally model system for perception-action ????


Why the fly ?

what makes a simple system


size, architecture, behaviour : no social system & communication, short life-span, fixed reflex chains ?

smaller brains generate simpler behaviour - at least smaller number of logical combinations
simpler to analyse - accessible to physiological study to identify the underlying neuronal circuitry 
>>>> fly behaviour is fast, robust, cheap


A simple system ?

comparison of size & complexity of two biological systems: man versus fly
rough estimates after Kirschfeld (1971), Braitenberg (1985)  

feature Homo sapiens Drosophila melanogaster Musca domestica factor
body size
70 kg
1.7 m
1 mg
3 mm
10 mg
6 mm
10,000,000
500
brain size
1.3 kg
0.1 m

1 mg
1 mm
1,000,000
100
number of neurons 10 10 - 10 11
10 6 100,000
eye design lens eye compound eye compound eye  
sampling points
(per eye)
10 6 700 3000 1,000

insects have radically different eye design: compound eye with hundreds of individual small lenses

<<<<<<

this eye design has a scaling problem: compound eye with the resolution of a human eye would look like this
(Kirschfed 1984)


Course stabilisation

perhaps the simplest task for a fly: stabilise a course against external disturbances (wind) and internal asymmetries (wing length, fatigue, load, … )


method to study this control system in the laboratory under idealised (easy to manipulate) conditions: torque meter and striped drums



yaw pitch roll

set of interconnected control systems: a versatile flight motor allows for three degrees of freedom – three axes of rotation yaw – pitch – roll (elaborate helicopter), see Blondeau & Heisenberg (1982)


Phenomenological model of feedback control

the analysis of course stabilisation in flies reveals a simple and clear structure of the underlying feedback control loop

(variant of general control circuit presented in lecture 1, the loop is opened in the experiments using the torque meter)

 

circuit to control flight torque and thrust/lift using motion information in Drosophila, from Goetz & Wenking (1973)


More advanced behaviour: orienting towards objects

flies exhibit exquisite aerobatics: fast and highly accurate flight manoevres, such as chasing in the behavioural context of mating


flight path of leading (open circles) and chasing (black dots) flies, from Land & Collett (1974)


free flight analysis reveals the statistics of object position on the retina of the chasing fly, which can be used to simulate flight paths


Object fixation in the laboratory

again open loop experiments can be designed using a torque meter to study the mechanisms of following a figure in front of a background
(and neuro-physiological implementation) under accessible & controlled conditions


(from Egelhaaf et al. 1988)

(from Reichardt 1986)


the fly is following (exhibiting a turning tendency towards = ‘tracking’) a single black stripe in front of a white background cylinder, or a textured stripe (figure) moving relative to a background of identical texture (ground)

the full characteristic for the turning tendency (torque) towards an object (D = direction-independent torque component) can be measured as function of object location  to describe the angular sensitivity of the feedback loop


Two optomotor systems

course stabilisation and object tracking can create substantial conflicts in the control of flight paths in natural environments: a turn towards a tracked object should be counteracted by the optomotor balance system!

how are the two optomotor responses coordinated ?
key observation: flies make very rapid turns when chasing objects

this behaviour corresponds to systematic laboratory measurements of torque in response to large-field motion (F&G) and small-field motion (F, located in the lateral field of view)

different responses have different dynamical properties: object response (small field) is tuned to fast image motion (4 Hz in the above experiment), course stabilisation (large field) is tuned to slow (0.0625 Hz) image motion (and therefore ‘blind’ to rapid turns during chasing)


The landing response



flies respond to image expansion (e.g. rotating spiral) with a characteristic landing response



the distribution of local preferred motion direction driving this response in fixed flight suggests integration of motion information matching expanding flowfields (Wehrhahn et al 1981)

the characteristic input pattern of motion signals and the dynamics of motion signal integration can be used to derive a phenomenological and computational model to describe the properties of the fixed flight landing response quantitatively



(from Borst 1990)

on the other hand, in free flight, flies approach landing objects with a characteristic speed profile using simple rules to estimate time-to-collision (Wagner 1982)
the properties of free flight landung dynamics resemble those of  human braking control, (see lecture 3)



free flying bees (in many aspects comparable to flies) land on textures disks elevated over a textured ground – they prefer to land on elevated disks and close to the boundary between disk and background – suggesting that speed differences (motion parallax) is used as critical cue to trigger the landing (Srinivasan 1992)


Speed control

further evidence that bees use image motion for flight control comes from experiments in which bees fly through tunnels with patterned walls



when the distance between the wall and the flightpath of the bee increases (decreases), the retinal pattern speed will decrease (increase)
– a control circuit stabilising image speed (similar to the optomotor equilibrium above) would predict that the bee should get faster (slow down)

this is exactly the pattern of results found for bees that were trained to fly through a double-cone shaped tunnel
(Srinivasan et al 1996)


similarly, the bee is expected to balance the image speeds on the two eyes, staying in the centre of a tunnel if both sidewalls are static, and de- or increasing the distance to a sidewall moving in the same or opposite direction as the bee, resp. (Srinivasan 1992)


this behaviuor is called 'centering response' - it is another consequence of  optomotor equilibrium !


Distance estimation

image speed is also used by bees to measure the distance travelled to a food source
which is then communicated to nest mates by means of the famous waggle dance




flying through a narrow tunnel of 6 m length is interpreted by the bee as flying a distance of almost 200 m through open landscape !!        (Srinivasan et al. 2000)


Landmark navigation


classical experiments by Tinbergen (1951) demonstrate that solitary wasps memorise the entrance of their nest by conspicuous landmarks (shifting the landmarks leads to search of nest entrance at a new location)

when leaving the nest, these wasps fly on characteristic flight paths (Collett & Lehrer 1993), looking back at the entrance from various locations – this is interpreted as ‘taking snapshots’, which could be interpreted as memorising a set of unique optic arrays in the framework of Gibson’s ecological approach

recent observations suggest that the highly regular oscillations on a virtual cone surface could also lead to a highly stereotypic pattern of flowfields, like dynamic landmarks…


Snapshot sequences

some ants, another group of hymenopteran insects, go on extended foraging excursions, and are known for highly sophisticated strategies to navigate back to their nests (sun compass, path integration, ...)

along large distances the pattern of available landmarks may change considerably, and it would be virtually impossible to home in on the target configuration of landmarks
- a recent study (Judd & Collett 1998) suggest that ants solve this problem by storing a sequence of snapshots which are later recalled in the right order to guide the way back


compare this type of representation to Gibson's notion of an optic array !


key reading:

comprehensive reference and reading list:

some study questions

download lecture handout


back to index page
last update 23/02/2004
Johannes M. Zanker