Rats are able to navigate in virtual environments
1 School of Biomedical Sciences, University of Ulster, Coleraine BT52 1SA,
Northern Ireland
2 Cognitive Neuroscience, Faculty of Biology, Tübingen University, Auf
der Morgenstelle 28, 72076 Tübingen, Germany
3 University of Freiburg, Chair of Pattern Recognition and Image Processing,
Georges Köhler Allee 52, 79110 Freiburg, Germany
* Author for correspondence (e-mail: hansjuergen.dahmen{at}uni-tuebingen.de)
Accepted 8 November 2004
![]() |
Summary |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Key words: virtual reality, spatial, memory, learning, rat
![]() |
Introduction |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
So far it has not been possible to create VR systems for rodents. Rodents
explore and navigate successfully in large areas. In a radio-tracking study,
home ranges of the rat Rattus rattus are reported to vary between
0.33 and 1.83 x 105 m2 and range lengths between
86 m and 311 m (Dowding and Murphy,
1994). In addition, much larger distances up to 954 m arereported
for migrating single recaptured males
(Hartley and Bishop, 1979
). In
contrast, investigations of the navigational abilities of rats have long been
restricted by the limited space of laboratoryoratories and the number of
landmarks that can be changed quickly and accurately without disturbing the
animal (Jeffery, 1998
).
Attempts have been made to construct a VE that could be used to investigate
the spatial navigation abilities of rodents. In one such set-up, animals were
trained in a Y-maze made of six monitor screens, and the scenes on the two
monitors that constitute an arm identified that particular arm. However, the
experimental results suggested that the animals did not treat the presentation
of scenes as VE in which to navigate, but rather as objects within the real
laboratoryoratory environment (Gaffan,
1998
; Gaffan and Eacott,
1997
; for a review, see
Hölscher, 2003
). We assume
that the reason for this result lies in the properties of the visual system of
the rat. The primate visual system has a relatively small visual field (about
the frontal half sphere), a fovea with color vision and high spatial two-point
resolution of about 0.015° (Zeki,
1993
). In contrast, the rat's visual field covers a large solid
angle (nearly the whole half sphere above and a large part of the half sphere
below the horizon (Hughes,
1977
). Rats also have a considerable binocular overlap. They do
not possess a fovea and have practically no color vision. They have a low
visual acuity for two-point discrimination of about 1° (Hughes,
1977
,
1979
;
Prusky et al., 2000
). The
rat's eye is adapted to low light intensities, the lens has a high
light-collecting power (f/d about 1;
Hughes, 1979
) and the retina
contains mostly rods (Szel and Röhlich, 1991). As a consequence, we
assume that any VR system for rodents must cover a large area of the visual
field but does not need high resolution and luminance.
We have developed a VR set-up that covers a large part of the rat's visual field (360° of azimuth, 20° to +60° of elevation). It is combined with a treadmill, in which the animal runs on top of an air-cushioned polystyrene sphere. Any translational movement of the animal leads to a rotation of the sphere, which is monitored and fed to the PC that controls the generation of the VE. The VE is rendered and presented to the animal in a closed action-perception loop. We developed this novel system to test whether or not rodents accept such virtual environments and can be trained to navigate in it in order to obtain food rewards. In the present paper we show that rats can indeed be accustomed to this VR system and successfully navigate in various environments for sugar water rewards.
![]() |
Materials and methods |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Apparatus
The treadmill
In our set-up, the rat runs in the zenith of a hollow polystyrene sphere of
50 cm diameter and a mass of about 400 g. The ideal mass of the sphere is 1.5
times the mass of the animal. In this case the haltered animal has to apply
the same force to accelerate the sphere as it would need to accelerate its
body mass by the same amount on the ground
(Dahmen, 1980). The sphere is
supported by an air cushion in a half-spherical mould of 50.4 cm diameter. The
animal is haltered by a soft leather harness attached to small aluminium
sheets that allow the rat to lift and lower its body, to pitch it slightly and
to rotate around the yaw axis. (Figs
1B,
2D). The animal rests with its
full weight on top of the ball. When the rat walks it stays in the zenith and
rotates the ball about a horizontal axis. The ball is prevented from rotation
around the vertical axis but can easily be rotated around any horizontal axis.
This is accomplished by slightly tilting the half-spherical mould (by about
7°). The polystyrene sphere is slightly pressed against two wheels (w in
Figs 1A,
2A), which touch the sphere
under 90° in its equator and are supported by horizontal axes with tip
bearings (Dahmen, 1980
). The
behaviour of the animal during an experiment is monitored by two small video
cameras (c in Fig. 2C). The yaw
angle (heading) of the animal is monitored by an angular incremental encoder
(aie in Figs 1A,
2D). In order to monitor the
locomotion of the animal, any motion of the ball is registered by two
x/y motion sensors (md in
Fig. 2B) positioned at 90°
along the equator of the sphere.
|
|
The sensors (HDNS2000 Agilent, Palo Alto, CA, USA) are motion detectors that are found in most optical PC-mouse devices. A statistical pattern of tiny black dots is applied with an airbrush onto the sphere surface. A small plastic lens (4.6 mm focal length) images this pattern, scaled down by a factor of 1/7, to the 1 mm x 1 mm light sensitive area (lsa) of the HDNS2000. The pattern is illuminated by a red LED, the intensity of which is chip-controlled. In this arrangement the ball surface may move to and away from the sensor by a few mm without degradation of movement detection. The maximum allowed speed of the surface is 2 m s-1 (30 cm s-1 on the lsa), the step resolution 0.5 mm (1/16 mm on the lsa). The signals of the detectors were fed to an incremental counter board (APCI1710, ADDI-Data, Ottersweier, Germany) implemented in the controlling PC.
The virtual environment
The rat is surrounded by a toroidal screen of 140 cm diameter and 80 cm
height (grey shaded in Fig.
1A). The screen covers the visual field from 20° below
to +60° above the horizon of the rat and 360° of azimuth (green shaded
in Fig. 1A). The image of the
VE is projected onto the screen from a DMD projector (b in Figs
1A,
2C) via two plane
mirrors (p) and an angular amplification mirror (aam). With respect to the
position of the animal and the aam the torus acts as a horopter: the animal
sees any object on the screen under the same angle as it is projected from the
aam (shaded blue in Fig. 1A).
The aam is a polished aluminium surface shaped as described by Chahl and
Srinivasan (1997;
Fig. 2C). It has a constant
`angular amplification' factor of 39 over its whole surface. (This
amplification factor is defined for two rays of the same azimuth as the ratio
of the divergence of the reflected beams to that of the incident ones.) Rays
that hit the aam far from its symmetry axis are reflected to the `sky' of the
rat, whereas rays incident near that axis are reflected to the `bottom' (red
shaded in Fig. 1A). The image
of the VE to be projected must be arranged in an annulus: the sky must be
morphed to the outer ring, the bottom to the inner ring of the annulus. This
morphing is done under the control of the graphics tool OpenGLPerformer (SGI,
Mountain View, CA, USA). A 360° panoramic view of the virtual landscape
seen from the momentary X/Y/Z position of the rat is rendered by 12
virtual cameras. This panorama is used as texture and applied to the annulus.
Another virtual camera grabs this textured annulus and produces the image for
the projector. Exploiting the hardware accelerated graphic commands of a
NVIDIA-GForce3 chip set, the above procedure is fast enough to synchronize to
the image repetition rate of the DMD projector.
Rewarding mechanism
In order to reward the animal, the haltering contains a thin brass tube
that feeds sugar water to the mouth of the animal
(Fig. 2D). A software triggered
valve controls the sugar water flow.
Experimental procedure
Because in VR it is difficult to simulate mechanical obstacles and because
we wanted to avoid conflicts between visual and tactile sensory input, we
designed an environment with cylinders suspended from the ceiling. Thus, the
animals could not encounter a visual obstacle that could not be sensed
mechanically. In order to be able to compare the learning abilities of rats in
virtual environments with those in real environments, rats were trained in a
real environment first in which there were real cylinders suspended from the
ceiling.
Experiment 1 in real environment
Six animals were trained in a laboratoryoratory to retrieve a cocoa
flavoured cereal (Kellogg's) located below one of three cylinders (20 cm
diameter, 50 cm in height). The cylinders were suspended from the ceiling
about 40 cm above the ground. One cylinder was white, one was black, and the
third one was of a grey colour. The cylinders formed a triangular
configuration and remained in their location throughout all runs. The rats
could explore a 2 m x 1.6 m area that was separated from the remaining
laboratoryoratory by white barriers 40 cm in height. Animals were given four
runs per day. The runs were recorded on video and analysed off-line. We
recorded the time until food reward was found, and the number of visits to a
20 cm catchment area below the three cylinders.
Experiment 2 in virtual environment
In our VE the ceiling was 1 m high and textured with a pattern of irregular
grey stones. The same texture was applied to the ground. The cylinders were 50
cm in diameter and 80 cm in height, suspended 20 cm above the ground. They
were covered with vertical black and white stripes. When the rat entered the
area below a cylinder it was rewarded with a drop of sugar water. In
preliminary tests some of the rats tended to return simply to a rewarded
cylinder to get another reward. We tried to prevent such `circling' around one
cylinder by defining an outer radius r0, which the animal
had to cross before it could get another reward at the same cylinder. The VE
was repetitive and was programmed as a grid of 5 x 5 identical squares.
The animal started in the middle one. When it reached a border of this inner
square the environment was shifted to the corresponding point on the opposite
border of this square. This quick change between two consecutive frames did
not change the appearance of the environment because of its periodicity. The
rendering of the scene was thus limited to the 25 squares. The animal's path
coordinates were restricted to the inner square while the rat was running for
hundreds of meters and the landscape looked endless to it. Afterwards the path
of the rat taken in the VE was reconstructed by eliminating the well-defined
trace jumps, which corresponded to border jumps in the registered data.
12 Long Evans rats were handled for 3 weeks and allowed to become accustomed to the harness, to movement on the polystyrene ball, and to the sugar water reward system. They were then trained in the above-mentioned square environment, which contained a square array of cylinders with 0.5 m diameter and 2m distance to each other. The above-mentioned r0 was 0.5 m. Each animal was allowed to run for 10 min on each of 10 consecutive days.
Experiment 3 in virtual environment
In another VR experiment starting 1 day later we presented a similar
`endless' square array of identical cylinders as in the previous experiment.
But now the distance between the cylinders had been increased to 10 m, and
r0 was 2 m. If the animal moved underneath a cylinder, the
visual angle subtended by each of the four nearest cylinders was about
4.6° of elevation and about 2.9° of azimuth. This implies that because
of the limited two-point resolution of the rat eye these cylinders are at the
limit of visual discrimination.
![]() |
Results |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
Experiment 2 in virtual environment
Results of experiments with cylinder spacing of 2 m
As shown in Fig. 4AC,
the animals improved considerably in their orientation to and in finding of
the cylinders. The numbers of rewards (hits per 10 min run) increased steadily
over the course of 10 days (one-way RM-ANOVA, F9,11=40.7,
P<0.0001, post-hoc Bonferroni multiple comparison test,
**P<0.01; ***P<0.001;
N=12; Fig. 4A). In
Fig. 4B the number of hits per
2 m distance increased over training sessions (ANOVA
F9,11=22.1, P<0.0001, post-hoc test
**P<0.01; ***P<0.001). While the
ideal path would have followed a straight line between cylinders (one reward
every 2 m), the animals improved to an average of 0.76 rewards per 2 m. The
path length per 10 min, i.e. the average speed of the animals, increased
(Fig. 4C). The maximum path
length per 10 min was 219.1 m. The average path length over all 12 animals
increased from 66.5 m to 166.2 m. Fig.
4D shows the inter-individual variation of the path length
averaged over all 10 days. As a general feature we observed that the animals
tended to follow a more-or-less defined direction in the laboratory coordinate
system over several days (Figs
5A,
7A). This direction varied from
rat to rat.
|
|
|
Fig. 5 demonstrates in more detail how the path chosen by the animals improved over time. In Fig. 5A the complete path of animal 9 for the first day (blue) and the tenth day (red) is displayed. The total path lengths were 70.6 m and 172.3 m, respectively. Fig. 5B,C depicts details of the paths boxed in Fig. 5A. Cylinders are drawn to scale as dark grey circles, hit cylinders marked by a larger light grey circular surround. Dots on the path mark every 30th sample (about every 1.4 s). Cylinder spacing is 2 m. The much more regular path on day 10 as compared to day 1 is apparent, and Fig. 5B,C depicts the most irregular part of the paths. The higher speed on day 10 can be inferred from the larger dot distance (Fig. 5B is at about twice the scale of Fig. 5C). The path on day 10 shows a more direct approach to the nearest cylinder. Fig. 5E,G shows the angular histogram of the orientation of the animal's body long axis, i.e. of the angular incremental counter (aie) data. If the animal were to run along its body long axis from one cylinder on a straight path to one of the four nearest cylinders, we would expect histogram peaks at N, E, S, W. On day 1 (Fig. 5E) we find no pronounced body orientation, while on day 10 (Fig. 5G) we find superimposed on a general orientation to SE, a peak orientation to S. In order to get an even better demonstration of the improvement of the `cylinder hit probability' of our rats, we surrounded each cylinder by a 2 m x 2 m square divided into 7 x 7 subsquares. We counted the number of trace samples in each subsquare and added up the counts in corresponding subsquares (corresponding with respect to their position relative to the central cylinder). Fig. 5D,F shows the relative number of counts in corresponding subsquares. For the trace of a rat that does not aim at the cylinders we expect an equal distribution of trace sample counts over all subsquares. The considerable increase of the `trace density' under cylinders from day 1 to day 10 is apparent.
Experiment 3 in virtual environment
Experienced rats from the previous experiment were transferred to a VE with
cylinder spacing of 10 m and r0 of 2 m. The results are
shown in Figs 6,
7, using the same conventions
as in Figs 4,
5. In
Fig. 6A the numbers of rewards
(hits) per 10 min run increased over the course of 5 days (one-way RM-ANOVA,
F4,11=7.3, P<0.0001, post-hoc
Bonferroni multiple comparison test, *P<0.05;
**P<0.01; ***P<0.001;
n=12). In Fig. 6B the
numbers of hits per 10 m distance run also increased over time (ANOVA,
F4,11=5.37, P<0.002, post-hoc
Bonferroni multiple comparison test, **P<0.01). The
maximum achievable number of hits per 10 m was 1; animals reached an average
value of 0.644. The average path length of 12 animals increased from 154.1 m
to 176.3 m per 10 min (Fig.
6C), the maximum path length was 228.8 m.
Fig. 6D shows the
inter-individual fluctuation of path length averaged over 5 days.
|
In Fig. 7A the whole trace of animal number 9, the same as in Fig. 5A, is displayed for day 1 (blue) and day 5 (red). The trace lengths are 123.7 m and 214.4 m, respectively. Fig. 7B,C show the boxed details of these traces. The traces appear, as compared to Fig. 5C, slightly less precise in their orientation to the nearest cylinders as demonstrated in the angular histograms of Fig. 7E,G. In Fig. 7D,E we surrounded each cylinder by a square of 10 m x 10 m and subdivided it into 7 x 7 subsquares. The relative trace sample density is very high in the subsquare under the cylinder (Fig. 7D,F).
![]() |
Discussion |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
While the good ability of rats to learn tasks in the VE is very encouraging, it should be noted that important sensory modalities such as smell or touch cannot be simulated. In addition, the virtual world is presented on a screen of fixed distance. Thus motion parallax is correct but stereo disparities are not correctly displayed. However, the importance of stereo vision to rats in navigation has not been demonstrated, and even if it is of some importance, the animals clearly adapted to the novel situation in which they were placed. We did not include acoustic characteristics of places, which might add to the realism of the rat's experience in VE. The vestibular stimuli that animals experience during ego-motion are correct for rotation but not for translation. Somatosensory force feedback to the legs because of rotational and linear acceleration and because of body weight are nearly correct. The fact that the animals did adapt to the unusual situation within the VE shows the degree of flexibility that rodents possess to adapt to unpredictable changes in their environment, and that they are able to ignore unreliable sensory input.
The time required and the distance run to obtain rewards continuously shortened during training, suggesting that the animals developed a strategy to optimise their foraging behaviour. The rat learned that running on the ball towards an object would bring the object `closer', perhaps by observing the image flow (e.g. the size of the 2-D projection of the object), which is the same as with real objects. We propose, as an explanation for our results, that the rats interpreted the 2-D projections as 3-D objects.
The strategies that the individual animals develop are diverse and range from simple running along rows of cylinders to circling strategies that increase the likelihood of finding a reward. Circling strategies were reduced by excluding an immediate second reward under the same cylinder. Most animals appeared to prefer a fixed running direction with respect to the laboratory coordinate system. This direction varied from rat to rat but was maintained over several sessions. Within the movements along that general direction, most rats clearly noticed the landmarks and navigated towards them. In the 2 m cylinder spacing experiment 2, a rat could easily see the four nearest cylinders from under the cylinder where it just had been rewarded. The cylinders covered an azimuth of about 14° and an elevation of about 22°. On day 10 many rats ran for a large part of their trace on a straight path from one cylinder to the next (see Fig. 5A,C and the angular distribution of the body length orientation Fig. 5G). The number of hits per 2 m trace length reached the average level of 0.76, which should be compared to the maximum possible value of 1. In contrast, in the 10 m cylinder spacing experiment 3, the nearest cylinders covering 4.6° of elevation and 2.9° of azimuth were not that easily visible to the rats. As a consequence, despite the training in experiment 1, traces are not so well oriented towards the nearest cylinders (see Figs 7AC). The animals did not quite reach the level of efficiency that they achieved in the experiments with 2 m cylinder distance. They hit fewer cylinders per 10 m trace length (0.64 compared to the possible maximum of 1) and the general direction of the trace dominated the angular histogram of body orientation (Fig. 7G). But when the cylinders became more visible, rats did turn towards them, find them and get a reward. The relative frequency of trace samples in the vicinity of cylinders shows a pronounced peak in the subsquare below the cylinder (see Fig. 7D,F).
The traces never became `optimal'. There seems to be always to be a tendency to deviate from the best path even when goals are clearly visible (see Fig. 5C). Perhaps we may interpret this `noise' as an investigatory strategy that cannot be suppressed. Since there are no repetitive and predictable natural environments with a 100% reward probability, it is of adaptive value constantly to deviate slightly from the seemingly perfect strategy in order to increase the likelihood of finding new food sources that otherwise would have been missed by running past them.
Conclusion
In conclusion, the results presented here show for the first time that rats
are capable of navigating in virtual environments, an ability that so far has
only been shown in humans and primates. The apparatus will open up new
possibilities such as the investigation of navigation in large spaces similar
in size to spaces that are explored by animals in the wild, and the importance
of specific sensory information such as visual landmarks, egomotion perception
from optic flow, or vestibular input in rodent orientation.
![]() |
Acknowledgments |
---|
![]() |
References |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Chahl, J. S. and Srinivasan, M. V. (1997). Reflective surfaces for panoramic imaging. Appl. Optics 36,405 -411.
Dahmen, H. (1980). A simple apparatus to investigate the orientation of walking insects. Experientia 36,685 -686.
Dowding, J. E. and Murphy, E. C. (1994). Ecology of ship rats (Rattus rattus) in a Kauri (Agathis australis) forest in Northland, New Zealand. NZ J. Ecol. 18,19 -28.
Gaffan, D. (1998). Idiothetic input into object-place configuration as the contribution to memory of the monkey and human hippocampus: a review. Exp. Brain Res. 123,201 -209.[CrossRef][Medline]
Gaffan, E. and Eacott, M. (1997). Spatial memory impairment in rats with fornix transection is not accompanied by a simple encoding deficit for directions of objects in visual space. Behav. Neurosci. 111,937 -954.[CrossRef][Medline]
Gillner, S. and Mallot, H. (1998). Navigation and acquisition of spatial knowledge in a virtual maze. J. Cog. Neurosci. 10,445 -463.[Abstract]
Hartley, D. J. and Bishop, J. A. (1979). Home range and movement in populations of Rattus norvegicus polymorphic for warfarin resistance. Biol. J. Linn. Soc. 12, 19-43
Hölscher, C. (2003). Time, space, and hippocampal functions. Rev. Neurosci. 14,253 -284.[Medline]
Hughes, A. (1977). The topography of vision in mammals. In Handbook of Sensory Physiology, vol.7 (ed. C. Crescitelli), pp.615 -756. Heidelberg: Springer Verlag.
Hughes, A. (1979). A schematic eye for the rat. Vision Res. 19,569 -588.[CrossRef][Medline]
Jeffery, K. J. (1998). Learning of landmark stability and instability by hippocampal place cells. Neuropharmacol. 37,677 -687.[CrossRef][Medline]
Lapointe, J. F. and Massicotte, P. (2003). Using VR to improve the performance of low-earth orbit space robot operations. Cyberpsychol. Behav. 6,545 -548.[CrossRef][Medline]
Leighty, K. A. and Fragaszy, D. M. (2003). Primates in cyberspace: using interactive computer tasks to study perception and action in nonhuman animals. Anim. Cogn. 6, 137-139.[CrossRef][Medline]
Liang, W. Y. and O'Grady, P. (2003). The internet and medical collaboratoryoration using virtual reality. Comput. Med. Imaging Graph 27,525 -534.[CrossRef][Medline]
Maguire, E., Frith, C., Burgess, N., Donnett, J. and O'Keefe, J. (1998). Knowing where things are: parahippocampal involvement in encoding object locations in virtual large-scale space. J. Cog. Neurosci. 10,61 -76.[Abstract]
Mishkin, M., Ungerleider, L. and Macko, K. (1983). Object vision and spatial vision: two cortical pathways. Trends Neurosci. 6,414 -417.[CrossRef]
Nishijo, H., Kazui, K., Hori, E., Tabuchi, E., Umeno, K., Sasaki, K. and Ono, T. (2003). Spatial correlates of monkey hippocampal neurons during navigation in a virtual space. Soc. Neurosci. Abstr. 32,717.14 .
Packard, M. G. and Knowlton, B. J. (2002). Learning and memory functions of the basal ganglia. Annu. Rev. Neurosci. 25,563 -593.[CrossRef][Medline]
Pollen, D. A. (1999). On the neural correlates
of visual perception. Cereb. Cortex
9, 4-19.
Prusky, G. T., West, P. W. and Douglas, R. M. (2000). Behavioral assessment of visual acuity in mice and rats. Vision Res. 40,2201 -2209.[CrossRef][Medline]
Rieser, J. J., Ashmead, D. H., Talor, C. R. and Youngquist, G. A. (1990). Visual perception and the guidance of locomotion without vision to previously seen targets. Perception 19,675 -689.[Medline]
Ruddle, R., Payne, S. and Jones, D. (1997). Navigating buildings in `desktop' virtual environments: Experimental investigations using extended navigational experience. J. Exp. Psychol. 3,143 -159.[CrossRef]
Szel, A. and Röhlich, P. (1992). Two cone types of rat retina detected by anti-visual pigment antibodies. Exp. Eye Res. 55,47 -52.[Medline]
Towers, D., Ellmore, T. and McNaughton, B. (2003). Spatial navigation by Rhesus monkeys in a virtual environment by exclusively visual cues. Soc. Neurosci. Abstr. 32,518.6 .
Wiener, J. and Mallot, H. (2003). `Fine-to-Coarse' route planning and navigation in regionalized environments. Spatial Cog. Comput. 3,331 -358.[CrossRef]
Zeki, S. (1993). A Vision Of The Brain. Oxford: Blackwell Scientific Publications.
Related articles in JEB: