Game of Drones: The plan to replicate the brain functions of bees within drones is in progress in the University of Sussex.
Biorobotics Professor Andy Philippides's team plans to design an autonomous flying robot with the navigational and learning abilities displayed by the highly intelligent forms of the bees. The four-year project called 'Brains on Board' involves researchers from the University of Sheffield and Queen Mary University of London. Commercial partners Nvidia have created small GPU chips (graphics cards) and tiny Jetson boards which the team will fit the unmanned aerial vehicles (UAV) onto. Postdoctoral researcher Alex Dewar told Reuters: "If you put additional weight on drones it makes them much more difficult to control, requiring more power and so on. The idea behind our project is learning from honeybees in particular and insects more generally, in terms of power-efficient strategies for carrying out autonomous behaviors."
Dewar says that creating such a computationally and energy-efficient autonomous robot would represent a step-change in robotics technology. He insists that despite their reputation as being highly effective team players, bees also show great individuality. "An individual forager is capable of going out and learning her environment, finding a patch of flowers, foraging, and coming home again and then communicating this information to other individuals. Bees have a remarkable intelligence in terms of navigation and also pattern recognition. Incorporating bees' neural capabilities in a UAV would potentially offer great improvements on typical drone control engineering solutions. Most other researchers interested in implementing intelligent behaviors on drones are more focused on deep learning. Philippides says the insects' brains provide an excellent autonomous system to reverse engineer, particularly as they pack so much power for their small size. He said: "Bees have various evolved behaviors - on a small scale called learning flights, at a large scale called survey flights - which they do innately. Our next stage is to look at how we can use behaviors we give the robot to enable it to structure the information coming in, so it makes the world easier to learn and easier to navigate."
The project aims to fuse computational and experimental neuroscience to develop a ground-breaking new class of highly efficient 'brain on board' robot controllers. These should exhibit adaptive behaviour while running on powerful, yet lightweight, General-Purpose Graphics Processing Unit (GPU) hardware now emerging in the mobile devices market. Autonomous and adaptive control of a flying robot, using an on-board computational simulation of the bee's neural circuits, would be an unprecedented achievement in robotics technology, says Philippides. He said: "What makes us unique is we've brought together experts in machine learning and computational neuroscience with roboticists and biologists. We've got a VR virtual reality arena where we can record what's going on in the neurons at very low scale. At the higher scale we have radar experiments, behaviour and neural recordings, plus the modelling expertise to put that together." The team will shortly test its newly-developed algorithms in a robot drone which they hope to have flying autonomously within a year.