Flying Insects and Robots

Insects use memorised visual representations to find their way back to places of interest, like food sources and nests. They acquire these visual memories during systematic learning flights or walks on their first departure and update them whenever approaches to the goal have been difficult. The fact that small insects are so good at localisation tasks with apparent ease has attracted the attention of engineers interested in developing and testing methods for visual navigation on mobile robots. We briefly review here (1) homing in insects; (2) what is known about the content of insect visual memories; (3) recent robotics advances in view-based homing; (4) conditions for view-based homing in natural environments and (5) issues concerning the acquisition of visual representations for homing. 7.1 Homing in Insects The ability of animals to recognise places of significance and to revisit them is fundamental to life on Earth. Without this navigational skill, for instance, many flowering plants would not be pollinated by insects and animals, in general, would be unable to provide for their offspring. There is ample evidence showing that animals including insects use memorised J. Zeil ( ) ARC Centre of Excellence in Vision Science and Centre for Visual Sciences, Research School of Biology, The Australian National University, Biology Place, Canberra, ACT 2601, Australia e-mail: [email protected] visual representations to pinpoint a goal location. The goal can be the nest location as in bees, wasps and ants (e.g. [2, 63, 68, 72], reviewed in [15]), the location of food as in bees (e.g. [9]) or hovering stations in flies and bees [16, 38, 39]. Visual spatial memories are crucial for local navigation but can also guide the animal during long-range navigation: routes can be formed from sequences of multiple stored views [14, 36]. Moving between these views, insects make use of compass and odometric information (see Chaps. 2 and 9). We will, in the following, focus on local homing methods that allow animals and robots to pinpoint a goal. It is clear since Tinbergen’s seminal experiments [63] that distinct objects in the vicinity of a goal location can act as landmarks and guide an insect’s return path. What constitutes a landmark under natural conditions, however, is still an open question. From a functional point of view, the following properties of objects are likely to make them useful as landmarks (see, e.g. [18, 26]): salience – landmarks should be unique and easy to distinguish from other parts in the scene; permanence or reliability – landmarks and their position should be constant over time; relevance – a landmark should help to recognise important places or decision points. However, when a homing agent has to acquire a visual representation of a place it wishes to return to it has – with the exception of salience – no obvious access to all these crucial, task-related properties of objects in the environment. How, for instance, is an insect or a robot to decide whether a particularly salient object it sees is permanent, reliable and relevant enough for the subsequent task of pinpointing that particular location? For the purpose of this chapter, we thus identify a number of open questions that should 87 D. Floreano et al. (eds.), Flying Insects and Robots, DOI 10.1007/978-3-540-89393-6_7, © Springer-Verlag Berlin Heidelberg 2009