Intelligence and Evolution
Survival of the smartest
Only one kind of device has ever displayed real intelligence: biological brains.
And only one kind of design paradigm has ever produced such a device: biological evolution.
Neuroblast creates artificially intelligent neural systems from a virtual evolutionary world of
genes, genomes, organisms and populations, all competing against one another in a rigorous selection
regime in which fitness is defined by one criterion: intelligence.
Our evolutionary world can engender almost any conceivable kind of neural processing system,
from simple neural networks to brains as complex as human cortices or even beyond.
Read on to see some of what they have achieved so far.
Brains for Robots
The coming revolution
The robotics industry is said to be right now where the PC industry was 30 years ago:
on the verge of exploding into a ubiquitous aspect of our everyday lives.
But one thing is holding this industry back: the robots do not have brains.
|
|
|
Neuroblast can evolve the brains for any type of robotic platform.
For example, the robot on the left, a Pololu 3pi
with five light sensors and two servo motors,
and controlled by one of Neuroblast's artificial brains, recently
finished in the top four at a regional robotics competition.
Not bad for our very first brain on our very first robot, but we are just getting started.
By controlling the selection regime, we can evolve brains to become ever more intelligent.
At present, we are selecting for appropriate responses to operant conditioning signals, the evolution of
which will allow users to train their robots to perform complex tasks just as easily as training a household pet.
|
Brains for Process Control
General-purpose no-tuning automatic control algorithms
The most common control algorithm in use today is the
PID controller.
Anyone who has ever used one knows how hard they can be to implement,
requiring the simultaneous optimization of two or sometimes three
dynamically-coupled tuning parameters.
Neuroblast has evolved drop-in alternatives that provide
much more power, with much less complexity.
|
|
|
The image on the left shows the disturbance-rejection performance of one
of Neuroblast's controllers (in red), compared to an expert-designed
PID controller (in grey). As usual, the Neuroblast controller was far more robust.
And yet while the PID controller was tuned for this specific process,
the Neuroblast controller was simple plug and play.
|
