Archive
Global and Local Scene Encoding
It is well understood that a key navigational mechanism for insects involves the learning of visual information from panoramic scenes. This leaves us with a basic question of how insects encode visual scenes for navigational. Computational studies have shown us how visual navigation can be achieved with: (i) Raw images; (ii) Sets of local visual features (such as oriented contrast edges), extracted from an image and tagged with retinal position; (iii) Sparse encodings where an entire scene is decomposed to simple parameters which represent a global property of the entire scene (such as centre of mass).
A life with the ants
It is almost impossible to believe that anybody looking at this page wouldn’t have heard of Ruediger Wehner and be aware of the pre-eminent role he has had in insect navigation research. In this biographical paper in the latest Ann Rev Entomol we get a lovely glimpse on the choices and situations that have shaped his career. All in all, it is a simple story of a research agenda driven by passion for insects that do remarkable things and thus demand our attention.
Wehner, R (2013) Life as a Cataglyphologist—and Beyond. Annual Review of Entomology. Vol. 58: 1-18
View-matching and image-matching
One of the significant contributions that studies of insect navigation can make to the wider field of comparative cognition is to suggest parsimonious mechanisms that might underpin spatial behaviour. One such mechanism is the use of egocentric views, which has been put forward recently as an alternative explanation to the presumed extraction of environmental geometry by vertebrates. In this article, we highlight how this debate (views vs geometry) can only be fruitful if we have a good sense of what information is available in an animal’s view. As an analytical shorthand, image matching is often used, however the important thing to remember is that animal’s views of the world will be filtered, processed and actively generated by movement therefore not always similar to a raw image. Understanding the information available in a view is essential before we can discard the use of egocentric views within a given experiment. For instance, the ability of vertebrates to obtain distance information and to use 3D views for navigation can be achieved egocentrically and included in view-based matching, even though it cannot not accounted by image-matching models.
Wystrach A, Graham P, 2012, “View-based matching can be more than image matching: The importance of considering an animal’s perspective” i-Perception 3(8) 547–549
Bayesian models of PI and search
This post is written by Rob Vickerstaff.
In our new paper Tobias and I present a Bayesian-statistical model of an ant searching for its nest using path integration (PI) as its only navigation aid, and compare the resulting search patterns to real Cataglyphis fortis search patterns and to three simpler search strategies. The ant search patterns were collected by Tobias, and we decided to use only searches by so-called zero-vector ants (those just emerging from their nest whose PI system was therefore known to be reset) because for these we knew there was minimal positional error in the PI system when the search began.
Our Bayesian search model might be called “semi-optimal” in that the model assumes a perfect memory of where search effort has been expended, but does not employ a sophisticated planning algorithm. Instead the ant is assumed to think only one step ahead, and choose the greatest immediate probability of finding the nest based on its current Bayesian probability distribution function. This simple rule is surprisingly robust, and outperforms the three simpler search methods in efficiency, and is the most ant-like in appearance.
A key feature of the model is that it automatically adapts to changes in positional uncertainty, a feature it shares with the ants which produce a broader search pattern the longer the preceding excursion has been.
Vickerstaff RJ and Merkle T (2012) Path integration mediated systematic search: A Bayesian model. J Theor Biol 307: 1-19
http://dx.doi.org/10.1016/j.jtbi.2012.04.034
Route navigation without waypoints
For an insect, the process of guiding oneself along a route is often assumed to be a matter of homing from one waypoint (defined by a snapshot) to another. However, this is a deceptively tricky process requiring decisions to be made about when and where to set waypoints. In this paper, we set out to demonstrate that visual route learning and guidance can be undertaken with a much more parsimonious architecture. Here is the author summary:
“The interest in insect navigation from diverse disciplines such as psychology and engineering is to a large extent because performance is achieved with such limited brain power. Desert ants are particularly impressive navigators, able to rapidly learn long, visually guided foraging routes. Their elegant behaviours provide inspiration to biomimetic engineers and for psychologists demonstrate the minimal mechanistic requirements for complex spatial behaviours. In this spirit, we have developed a parsimonious model of route navigation that captures many of the known properties of ants routes. Our model uses a neural network trained with the visual scenes experienced along a route to assess the familiarity of any view. Subsequent route navigation involves a simple behavioural routine, in which the simulated ant scans the world and moves in the most familiar direction, as determined by the network. The algorithm exhibits both place-search and route navigation using the same mechanism. Crucially, in our model it is not necessary to specify when or what to learn, nor separate routes into sequences of waypoints; thereby providing proof of concept that route navigation can be achieved without these elements. As such, we believe it represents the only detailed and complete model of insect route guidance to date.”
Baddeley B , Graham P , Husbands P , Philippides A , 2012 A Model of Ant Route Navigation Driven by Scene Familiarity. PLoS Comput Biol8(1): e1002336.
The role of ocelli in ant navigation
Re-posted as paper now published.
Revised post with author comments:
As well as their compound eyes, many insects possess light sensitive ocelli, which usually come in threes and are positioned on the top of the head. Their role in the stabilization of flight is relatively well-studied but little is known about the use of ocelli by walking insects. It was shown by Wehner for Cataglyphis that ants can use their ocelli to extract celestial compass information. Schwarz et al have recently confirmed this is also true for Melophorus and in this paper they extend that work to look at whether compass information extracted by the ocelli can be used by the path integration system. They found that ants with ocelli (but compound eyes covered) were unable to do standard path integration – suggesting compass information derived from eyes and ocelli are not ‘merged’. Instead, the ants with only ocelli were only able to aim in a direction opposite to the most recent leg of their journey. The role for this is not clear.
Sebastian Schwarz explains their study:
“After a foraging trip, zero-vector ants, without any information from their path integration system and covered compound eyes but open ocelli, headed opposite to the direction they came from. This unexpected but nonetheless interesting behaviour led us to a more detailed examination of the interaction of ocelli and path integration. We found that ocelli alone cannot mediate path integration like compound eyes do. After a two-legged outbound route, ant with covered compound eye but functional ocelli did not head towards their nest, as the path integrator would indicate, but headed instead towards the last leg of travel. The ocelli mediated compass might be a back up system in case the foraging ant was forced to leave her familiar foraging route and thus, supports a more effective homing in their visually cluttered environment. Ocelli were known for their function in flight stabilisation, but in ground-based insects their role in such a distinct compass system raises a number of functional, mechanistic and evolutionary questions.”
Sebastian Schwarz, Antoine Wystrach, and Ken Cheng (2011) A new navigational mechanism mediated by ant ocelli. Biol Lett published 6 July 2011, 10.1098/rsbl.2011.0489
Holistic encoding of visually guided routes
I’m not sure how I managed to miss this paper for the BLOG – given i’m a co-author – but better late than never – pg
The inspiration for this project was to try and capture in an artificial algorithm some of the robust performance seen in the long visually guided routes of ants, as the mechanisms by which visual information is first learned and then used to control a route direction are not well understood. In this article, we propose a parsimonious mechanism for visually guided route following. We investigate whether a simple approach, involving scanning the environment and moving in the direction that appears most familiar, can provide a model of visually guided route learning in ants. We implement view familiarity as a means of navigation by training a classifier to determine whether a given view is part of a route and using the confidence in this classification as a proxy for familiarity. Through the coupling of movement and viewing direction, a familiar view specifies a familiar direction of viewing and thus a familiar movement to make. We show the feasibility of our approach as a model of ant-like route acquisition by learning a series of nontrivial routes through an indoor environment using a large gantry robot equipped with a panoramic camera.
Bart Baddeley, Paul Graham, Andrew Philippides, and Philip Husbands (2011) Holistic visual encoding of ant-like routes: Navigation without waypoints. Adaptive Behavior 19: 3-15
Searching for your nest
One of the cornerstone mechanisms for insect navigators is a systematic search strategy, for instance used when looking for their nest after their path integration system vector has “zeroed out”. The beauty of these searches is that they aren’t fixed motor patterns, rather, they reflect an individual’s confidence in its spatial knowledge. In this paper, Schulthiess and Cheng, study the statistical details of the nest search of the Australian desert ant.
Patrick Schultheiss writes “ When foragers of the Australian desert ant Melophorus bagoti navigate home, they use both visual navigation mechanisms and path integration. If they reach the nest area but fail to find the entrance, they engage in a systematic search. We looked at this search in displaced foragers, after they had completed the inbound journey in their familiar environment. It is made up of search loops, is centred on the (fictive) nest location and gradually increases in size. Similar to Tunisian Cataglyphis ants, a longer inbound journey leads to a larger search, most likely due to a decrease in accuracy of the homing vector. The interesting thing is that this effect is displayed after the ants navigated home in their familiar environment; so the (more accurate) process of visual navigation did not eliminate the (cumulative) errors of the path integrator.
We also investigated the search strategies used by Melophorus foragers. In optimal searching theory, animals are considered to move in straight lines (segments), separated by incidents of reorientation. In different search models, the segment lengths are drawn from different mathematical distributions (e.g. exponential, power law). We find that our searches are best described by a double exponential function. What this actually means is that, in the unfamiliar test-field, our ants show a mixture of two searches: one search with short segments, and one with long segments. Short segments may be useful for tight searches around the (fictive) nest, and long segments may be good to ‘loosen up’ this search and investigate new areas.”
Schultheiss, P. & Cheng, K. (2011): Finding the nest: Inbound searching behaviour in the Australian desert ant Melophorus bagoti. Animal Behaviour, doi: 10.1016/j.anbehav.2011.02.008
A new piece of the Melophorus jigsaw
I have re-posted this entry to reflect the fact that the pdf is now available and to correct the citation.
Because of its interestingly cluttered environment, the Australian desert ant Melophorus bagoti has become a really useful study species for navigation research. In particular their use of visual cues. In this paper, we add another piece to this story by looking at the structure of Melophorus’ compound eyes. The ants possess typical apposition eyes with about 500 ommatidia per eye and a horizontal visual field of approximately 150°. The average interommatidial angle
Δϕ is 3.7°, the average acceptance angle of the rhabdom Δρ(rh) is 2.9°. With a Δρ(rh)/Δϕ ratio of much less than 2, the eyes undersample the visual scene but provide high contrast, and surprising detail of the landmark panorama that has been shown to be used for navigation – Sebastian Schwarz
Schwarz S, Narendra A, Zeil J. (2011) The properties of the visual system in the Australian desert ant Melophorus bagoti. Arthropod Structure & Development 40 (2011) 128-134
Do ants need discrete snapshots?
It is well established in the literature that a single view of the world is unique to the location from where the view was perceived/stored. This leads to the idea of snapshot guidance to a single goal. A logical extension is to imagine routes could be composed of multiple snapshots which an ant uses in a sequence – an idea which was strongly suggested by specific experimental evidence. Here we present modelling results which show how evidence that had
suggested the use of discrete snapshots can in fact be produced with a mechanism that does not use discrete snapshots.
Antoine Wystrach, Michael Mangan, Andrew Philippides and Paul Graham (2013) Snapshots in ants? New interpretations of paradigmatic experiments. J Exp Biol 216:1766-1770