Using reconstructed visual reality in ant navigation research

Abstract

Insects have low resolution eyes and a tiny brain, yet they continuously solve very complex navigational problems; an ability that underpins fundamental biological processes such as pollination and parental care. Understanding the methods they employ would have profound impact on the fields of machine vision and robotics. As our knowledge on insect navigation grows, our physical, physiological and neural models get more complex and detailed. To test these models we need to perform increasingly sophisticated experiments. Evolution has optimised the animals to operate in their natural environment. To probe the fine details of the methods they utilise we need to use natural visual scenery which, for experimental purposes, we must be able to manipulate arbitrarily. Performing physiological experiments on insects outside the laboratory is not practical and our ability to modify the natural scenery for outdoor behavioural experiments is very limited. The solution is reconstructed visual reality, a projector that can present the visual aspect of the natural environment to the animal with high fidelity, taking the peculiarities of insect vision into account. While projectors have been used in insect research before, during my candidature I designed and built a projector specifically tuned to insect vision. To allow the ant to experience a full panoramic view, the projector completely surrounds her. The device (Antarium) is a polyhedral approximation of a sphere. It contains 20 thousand pixels made out of light emitting diodes (LEDs) that match the spectral sensitivity of Myrmecia. Insects have a much higher fusion frequency limit than humans, therefore the device has a very high flicker frequency (9kHz) and also a high frame rate (190fps). In the Antarium the animal is placed in the centre of the projector on a trackball. To test the trackball and to collect reference data, outdoor experiments were performed where ants were captured, tethered and placed on the trackball. The apparatus with the ant on it was then placed at certain locations relative to the nest and the foraging tree and the movements of the animal on the ball were recorded and analysed. The outdoor experiments proved that the trackball was well suited for our ants, and also provided the baseline behaviour reference for the subsequent Antarium experiments. To assess the Antarium, the natural habitat of the experimental animals was recreated as a 3-dimensional model. That model was then projected for the ants and their movements on the trackball was recorded, just like in the outdoor experiments Initial feasibility tests were performed by projecting a static image, which matches what the animals experienced during the outdoor experiments. To assess whether the ant was orienting herself relative to the scene we rotated the projected scene around her and her response monitored. Statistical methods were used to compare the outdoor and in-Antarium behaviour. The results proved that the concept was solid, but they also uncovered several shortcomings of the Antarium. Nevertheless, even with its limitations the Antarium was used to perform experiments that would be very hard to do in a real environment. In one experiment the foraging tree was repositioned in or deleted from the scene to see whether the animals go to where the tree is or where by their knowledge it should be. The results suggest the latter but the absence or altered location of the foraging tree certainly had a significant effect on the animals. In another experiment the scene, including the sky, were re-coloured to see whether colour plays a significant role in navigation. Results indicate that even very small amount of UV information statistically significantly improves the navigation of the animals. To rectify the device limitations discovered during the experiments a new, improved projector was designed and is currently being built.