This Drone Has Artificial Intelligence Modelled on Honey Bee Brains
Forget merging man and machine, these researchers are fusing bee and drone.
Image: Alex Cope/Green Brain Project
Artificial intelligence aficionados such as Singularity-chasing Ray Kurzweil might be working on fusing together man and machine, but scientists in the UK have started on a smaller scale, with the goal of meshing together bee and drone.
Launched in 2012, the Green Brain Project aims to create the first accurate computer model of a honey bee brain, and transplant that onto a UAV.
The project, based out of the University of Sheffield and University of Sussex, seeks to raise awareness of the declining population of honey bees worldwide, as well as to advance our knowledge of AI and honey bee cognition.
While bees have 960,000 neurons, compared to the 100 millions neurons in a human brain, scientists have discovered that a honey bee's brain is in fact impressively responsive and adaptive.
Researchers from the Green Brain Project—which recalls IBM's Blue Brain Project to build a virtual human brain—hope that a UAV equipped with elements of a honey bee's super-sight and smell will have applications in everything from disaster zone search and rescue missions to agriculture.
"Bees have tremendous navigation abilities, and are able to learn and follow routes up to around seven miles,"said James Marshall, a professor of theoretical and computational biology who is heading up the project.
So how close are we to having our first bee brain-powered UAV? Marshall said that the team had made significant progress since 2012 and are now at the stage where they're running bee brain simulations on their own bespoke quadcopter platform. "We are able to mount a variety of sensors on this quadcopter to mimic the sensory ability of the honey bee," he said in an email.
The team is replicating the honey bee's wide field of vision with two fish-eye cameras, and mounting chemosensors, which act like "electronic noses" that detect odorants with particular chemical profiles.
As this UAV will be unleashed into the air in the future, progress has been made in modelling the honey bee's visual processing system. Honey bees are known to calculate angular velocity and use this to avoid flying into walls and to regulate flight speed, Marshall explained.
In this video, the drone regulates its distance from the wall using visual information
Modelling any organism's brain is no easy feat, and reproducing what happens in a computer simulation on a physical robot always proves a challenge. "Using models that run well in virtual environments usually throws up unanticipated problems when deployed in the real world as it's so much richer and variable than even the most advanced virtual environment," said Marshall. He pointed out that algorithms that work well in theory could only be trusted when they've been proven to work in practice in the "real messy world" that we live in.
Since 2012, Marshall said the team had successfully closed the loop between the honey bee brain simulations that they made on computers and the flying UAV, "with sensory input coming into the brain simulation from the robot, and motor commands being issued by the brain simulation and implemented by the flying robot."
But what makes the honey bee's brain such a desirable starting point for an enhanced UAV? Honey bees, or Apis mellifera, have been long studied owing to their sensory and visual capacities and strength in navigation, communication, and learning and memory processing skills.
In 2013, animal cognition researchers, Aurore Avarguès-Weber and Martin Giurfa conducted a study that showed honey bees were capable of learning "concepts." The experiment revealed how—despite their tiny brains—honey bees could get to grips with conceptual relationships such as "similarity and difference," "above and below," and "left and right" that depend on associations between objects rather than just the physical properties of objects.
Marshall described how for this experiment honey bees were trained to associate a stimulus presented at a maze entrance with the same stimulus providing a reward within the maze.
"This kind of concept learning is one of the abilities we'd eventually like to be able to replicate in our models," said Marshall, who noted that a robot control system as autonomous and adaptive as a bee would have great potential.
As the world's pollinators, bees possess top-notch foraging skills. "They are good at search, discovery and reporting locations," he said. "If we can eventually adapt these kind of abilities to the exploration of hostile, unfamiliar environments such as search and rescue scenarios for example, that could be a real breakthrough."