MiRo was created to sense the world and then act on it. In this two part blog post, we are asked Consequential Robotics CTO Dr Ben Mitchinson to talk us through the origins of MIRO’s biomimetic brain. Mitch (Ben) has spent the last fifteen years studying both animals and automatic control systems. His background is the physical sciences and engineering, within which he has broad experience of all things artificial. In software alone, he is familiar with more than twenty programming languages (and has written a couple). His passion is to bring those artificial systems to life - seeking links between biological and human-designed control systems and morphologies, and taking advantage of those links both to advance our understanding of animals and, at the same time, to begin to breathe life into our machines.
In part one, Mitch will introduce us to the thinking behind making models and brain models!
Models and brain models
A technique that scientists and engineers use when trying to understand what they call ‘complex systems’ is to build ‘models’ of those systems which they can then manipulate and adjust to find out how they behave under different conditions. Models can take many forms - architects and civil engineers, for example, build actual physical models, scaled-down, of their projects. Neuroscientists aim to understand data processing functions in the brain, so physical models won’t do - instead, they build mathematical models of those data processing functions in a computer.
It is worthwhile understanding how a bridge will behave before you actually build it.
The Tacoma Narrows Bridge Collapse (1940).
https://en.wikipedia.org/
A key contribution that these models make is that they shake out misunderstandings. If the scientist or engineer doesn’t understand some part of the system they are modelling as well as they thought, they will find out that it doesn’t ‘work’ when they model it. If the model bridge falls down, then it is not a good model of a (working) bridge - analogously, if the model brain does not, in some sense, function like a real brain then it is not a good model of a brain. At this point, it’s obligatory to quote Richard Feynman, who wrote on this subject that “What I cannot create, I do not understand”.
Models that do not work, however, are not useless. The engineer might well learn what is wrong with their bridge design from observing the collapse of the model. Similarly, a brain model that does not function like a brain can still teach the scientist something - examining a model that is not working correctly can help to generate new hypotheses as to how it might be constructed, hypotheses which can then be tested against a real brain, just like model bridges are ultimately tested by turning them into real bridges.
Scientists and philosophers have proposed many models of the brain over the centuries
Phrenology is an early model of brain function that today is considered to be rubbish.
https://en.wikipedia.org/
Early models were based on nothing much more than intuition, there being no way to measure the operation of the brain at that time, and most of them are no longer given much credit. In recent years, there have been enormous advances in our ability to measure what the brain is doing at a detailed level (recordings can be taken from individual neurons, moment by moment). At the same time, computers have become available that are fast and user-friendly - brain models have become far more detailed, and more useful, as a result of these developments.
The scientific community is still far from agreeing on how best to understand the brain, however, and what makes a good and useful model remains hotly debated. Our research group in Sheffield is involved in this debate, and has been building models of parts of brains for many years.
How do we test our models to see if they ‘work’ like a real brain?
One answer is that brains are for controlling behaviour, so if a brain model is a faithful model of a real brain, we should be able to use it to generate sensible behaviour from something like an animal body. That is why a key part of our work in Sheffield is to build animal-like robots that use model brains as controllers, and test whether they behave as we would expect when challenged with simple tasks, like navigation or decision-making. For the science team in Sheffield, MIRO is the latest in our line of research robots with a ‘biomimetic’ brain.
Stay tuned for Part 2 to learn about MiRo’s brain model!