The findings shed light on how brains resolve one of the most difficult challenges of vision � namely, recognizing different surfaces under different colours of illumination � by suggesting that bees solve this problem using their experience with meaningful colour relationships between objects in a scene. The findings, published in the Proceedings of the National Academy of Sciences, may one day lead to the design of autonomous robotic systems.

In the UCL study, scientists from the UCL Institute of Ophthalmology trained bumblebees to find artificial flowers of a particular colour using a nectar reward. They then tested the bees' ability to find the same flowers in scenes that were simultaneously illuminated by four differently coloured lights � UV-yellow, blue, yellow and green. To solve this puzzle, the bees had to effectively segment the scene into its different regions of illumination, and then find the correct flowers within each region.

Dr Beau Lotto of the UCL Institute of Ophthalmology says: "Although we knew that bees were able to recognise flowers under different global lights, we didn't know whether they could also do this under more complicated conditions, ones that are in fact more typical in nature, such as dappled light across a woodland floor.

"When all the surfaces in a scene are under the same light, identifying a particular surface when the global illumination changes is in principle an easy problem to solve, since all vision needs to do is adapt itself to the scene's average colour, a bit like adapting to the darkness of a cinema. "Far more difficult is to recognise the surface or object under multiple lights simultaneously, since adapting to the scene's average colour � which was previously thought to be the strategy used by bees � won't work."

"Our study shows that the tiny brain of the bee can not only solve this difficult task, which the most sophisticated computers still can't resolve, but suggests they do so by using the colour relationships between objects in a scene that were statistically most useful in their past experience. Because this same strategy is also used by humans, our work on bees, in conjunction with our work on humans, may enable us to understand the general principles by which any visual system (natural or artificial) can construct useful behaviour from ambiguous sensory information.

'One long-term aim of our research is to exploit this understanding to build seeing robots that, like the bee with its mere one million neurons, can learn to find a simple flower in a meadow, which no machine can do at present. Our lab has reconstructed our specially designed bee flight arena � known as the Bee Matrix � in the virtual world, where virtual autonomous bees are 'evolving' under exactly the same conditions as those experienced by our real bees."

Related Links
SpaceDaily
Search SpaceDaily
Subscribe To SpaceDaily Express