Route learning by insects Thomas S Collett  , Paul Graham and Virginie Durier Ants and other insects often follow fixed routes from their nest to a foraging site. The shape of an ant’s route is set, initially, by navigational strategies, such as path integration and the ant’s innate responses to landmarks, which depend minimally on memory. With increasing experience, these early routes are stabilised through the learning of views of landmarks and of associated actions. The substitution of memory-based strategies makes an insect’s route more robust and precise. The ability to select between different learnt routes might incur additional memory requirements to those needed for performing a route, and lead to the associative grouping of those memories that relate to a particular route. Addresses Sussex Centre for Neuroscience, University of Sussex, Brighton BN1 9QG, UK  e-mail: t.s.collett@sussex.ac.uk Current Opinion in Neurobiology 2003, 13:718–725 This review comes from a themed issue on Neurobiology of behaviour Edited by Mark Konishi and Randolf Menzel 0959-4388/$ – see front matter ß 2003 Elsevier Ltd. All rights reserved. DOI 10.1016/j.conb.2003.10.004 Abbreviations PI path integration Introduction Learning of a route seems to be a demanding task for an insect nervous system. Yet, some insects, particularly ants and bees, readily acquire memories of landmarks and of actions associated with those landmarks, and use those memories when navigating between their nests and a foraging area (reviewed in [1,2  ]). These routes can be extraordinarily long; the record is probably held by orchid bees – they are reputed to trap-line over 10 km and visit many sparsely distributed orchids in a fixed order [3]. Such extravagant routes are hard to monitor and, as a result, our current knowledge of route acquisition comes mostly from earthbound insects, mainly ants. Although it is often presumed that ants merely follow chemical trails, in reality, their navigational strategies are far more diverse. Desert ants, for example, forage indi- vidually and do not lay chemical trails at all; nevertheless, they perform exactly the same route over several trips [4–6]. Even the foragers of some ant species that do lay chemical trails rapidly learn to guide themselves by the visual landmarks that they have viewed from the trail [7,8]. The trail provides a scaffold upon which visually controlled routes can be assembled; when, in experi- ments, chemical trails and visual landmarks are placed in conflict, experienced ants are directed by the visual landmarks, whereas naı¨ve ants follow the chemical signals [8]. This switch reflects a common feature of route acquisition: the route that an ant first takes is determined by early strategies that require little or no learning and which remain available as a backup. With experience, the insect can adopt more-efficient, memory-dependent stra- tegies. In the present example, ants travel faster and are in less danger of losing the route when they use learnt visual cues, than they are when reliant on chemical trails. In this review, we focus principally on ants and concen- trate on four topics related to the learning of visually guided routes: (1) scaffolds for route acquisition, (2) the use of snapshots in route guidance, (3) where and when to learn, and (4) learning speed. Scaffolds for route acquisition: path integration and innate responses to landmarks Without the benefit of a chemical trail, an inexperienced desert ant at the start of its foraging career relies on path integration (PI) to elaborate a route [6]. It monitors the direction and distance of its outward path, integrating this information to provide an up-to-date estimate of its current distance and direction from the nest. When it comes across a food item, it stores the coordinates of the site in terms of the state of its PI system and then uses PI to return home. The stored coordinates of the food site then set the destination of the ant’s subsequent food- bound trip [9,10]. Aside from calibrating the sun compass [11,12] and possibly other components [13,14  ], PI requires no long-term memory, and can, therefore, guide the routes of inexperienced ants. Desert ants that primar- ily collect dead insects tend to explore outwards from their previous feeding site [9,10]; by returning to previously visited places, ants repeatedly move through the same terrain, and give themselves the opportunity to learn routes through it. With increasing experience, the ants’ foraging routes tend to get longer but remain within a fixed sector, centred on the nest [10]. This outward progression leaves prey nearer to the nest for younger foragers to find. Ants navigating by PI, thus, learn the appearance of visual landmarks that are viewed en route. Subsequently, they can perform a route that is emancipated from the PI, which initially provided a scaffold for learning. In the example of Figure 1, a barrier was placed perpendicular to 718 Current Opinion in Neurobiology 2003, 13:718–725 www.current-opinion.com