The Green Brain team worked to build a neuromimetic model of the honeybee brain and use it to control a flying robot, with an aim to better understand the way that the honeybee brain works.
We can apply our understanding of how the bee brain functions in several ways:
- to design and develop better algorithms for guiding autonomous flying vehicles
- to provide insight into how all brains work by translating the lessons we learn from the less complex bee brain to the brains of mammals
- to understand the neurophysiological basis of the threats that these essential pollinators face in the modern world.
Visit our Publications page for available software and publications, or check our Contact Us page for links to the BrainsOnBoard YouTube channel and GreenBrain Facebook page.
Generally speaking, the necessary objectives to achieve this goal included modelling the brain, building the robotic platforms, and developing the toolchains that allow the model to run in real-time on GPUs.
At this stage in the project, our honeybee brain models consisted of the fundamental processes involved in vision and olfaction, and how these relate to behaviour. The brains models were then further developed and tested using the Green Brain robot team.
The below links will explain more about the different areas of the project:
The Green Brain team aimed to make a neurologically based model of the honeybee brain and embody it within robotic platforms.
Neural systems can be modelled at various levels of detail – from the movement of ions across cell membranes to treating groups of neurons as single computational devices. The Green Brain team used an approach where we modelled individual neurons but treated these as having no spatial extent.
To simulate how these individual neurons react to their inputs, we used the simplified Izhikevich formulation or simple leaky integrator equations for the membrane dynamics. The behaviour of the models was therefore largely determined by the way that the neurons are interconnected.
We aimed to build closed-loop systems: modelling complete neural circuits from the sensory input to the motor output. Closed-loop modelling allows behavioural data from bees to be used in testing the models. This provided a richer set of data to explore how our models performed, as well as being essential for embodying our models in our robots.
Behavioural experiments on honeybees have a long history. One of the best tools for vision, olfaction, and decision-making testing is the Y-maze.
The Y-maze task
Via the Y-maze, bees can be tested abilities like speed-accuracy trade-offs, positive and negative conditioning, and concept learning and transference. Among various other testings, the ultimate assessment of the Green Brain will be evaluated on its completion of this task in the lab.
- Bees enter through the front (bottom of diagram) and are faced with two maze arms.
- At the decision point, the two stimuli subtend a fixed visual angle.
- The bee must enter the correct arm of the maze in order to receive a reward.