From robot dreams to reality
One of the most ambitious goals of artificial intelligence
Life research is the construction of life system from scratch. living parts.
Most artificial systems still exist strictly inside the computer, but on this issue liplipson and Jordan Pollack take the first step towards bridging the gap between the computer model and the physical reality.
They describe a system that develops locomotive machines in computers and then uses rapid generation
So they can move around in the physical world.
In the past 20 years, computer algorithms inspired by genetics (
Has become a common tool to solve optimization problems.
Typically, the optimization objective is a mathematical process in a known form, but there are many parameters to be determined.
The overall candidate process is generated by representing different sets of these parameters as binary strings.
Evaluate the performance of each program, allowing reproduction of programs that are \"appropriate\" or do better according to some specific criteria.
Replication can be sexual by taking the strings of two fit parent procedures and using cross-to generate binary strings for future generations
It may or may not be sexual, in which case only mutations are used.
In many generations, people tend to program more successfully.
Over the past decade, many people have tried the evolving \"artificial\" population in a simulated environment or in a virtual world.
Here, the string is not a binary string representing certain process parameters, but rather an artificial genome encoding the control circuit (
Or nervous system)
A simulated robot.
So, over time,
The acting robot appears slowly.
In some cases, there is an implicit adaptive assessment when a creature struggles for the virtual resources necessary to survive.
In other cases, each generation of organisms has a clear \"adaptation function\" that forces evolution to move in the desired direction.
There are also some experiments in transferring robot control systems that evolved inside the computer to physical versions of analog robots.
But there is a lot of debate and conflicting evidence about the effects of these transfer experiments.
In some cases, for dozens to hundreds of generations of 10,000 people, about 100 of physical fitness assessments are done on physical robots --
But it is understandable that this experimental tenacity is rare.
One obvious omission in these experiments is that the robot\'s \"body\" is often considered constant, while only the nervous system is evolving.
Carl Sims\'s work is an encouraging exception.
He evolved creatures that could swim and crawl in a virtual world that follows Newton\'s laws of physics.
His fitness function rewards horizontal sports and outputs pop-up creatures with locomotive capabilities, although some adjustments are needed to work as planned.
The earlier version of the fitness function did not punish the vertical movement, only for a few seconds of existence, so the really tall creatures evolved creatures that were good at falling down or even tumbling.
For a while, creatures evolved by beating the body with limbs --
They took advantage of a defect in simulated physics that encoded conservation of momentum.
Sims\'s system is a co-evolution of the nervous system and the physical plan, but his creatures are purely calculated.
A few years ago, Paul Forness and Jordan Polak tried to convert computer models into physical reality by evolution rather than biology, just structure
The simulated structure is selected according to the strength.
They then hand-built physical versions of these structures with real Lego blocks and confirmed that these structures are much stronger than humans --designed ones.
Lipson and Polak are now taking the idea one step further.
They develop locomotive systems in computing space and use fast
Prototype technology for automatic production of multiple products
Only the link structure of the motor needs to be broken manually.
Successful Design evolved surprisingly different modes of motion production, but there was a reasonable correlation between the model\'s predictive locomotive capability and the physical robot\'s measurement capability ().
Body parts and control circuits evolved within the computer.
Thousands of generations of robotsand then rapid-
Prototype technologies are used to turn them into reality.
Some of the award-winning designs are surprisingly symmetrical, which may be easier to explain for symmetrical machines to find linear motion.
This particular machine is moved using inverse bit sync
When the upper limb pushes the machine forward, the central body recovers and vice versa.
The movie for this robot and others is available atas supplementary information.
Lipsen and Polak\'s experiments were carried out on their computers.
There is no fitness assessment in the physical world.
The computing part of the robot is left in the same computer even if it has been physically constructed.
This means that there can be no feedback from the physical world during evolution.
At best, this system is like a virus that uses other more complex machines (
This is not life in this case
There is still a way to go before self.
Replication robots can exist in the real world.
But for artificial
Life researchers have some aspects of their current research that resonate satisfactorily with the way life systems are developed.
First of all, the traditional manufacturing technology can not make these special robots. The rapid-
The prototype technology solidified the polymer in place so that the ball and socket joint could be constructed with a ball inside the socket.
Parts will never be separated, and if they are separated, they cannot be assembled without damaging them.
This is not far from the way the biological system grows.
Second, the evolutionary strategies used in these experiments, starting with a blank or \"empty\" genome, randomly mutates it into a genome that produces a working machine --
So there\'s no-
Build prejudice from seed machines in the population.
It can be said that these machines evolved \"naturally\" without human intervention.
People have long wanted
Biological materials capable of self-breeding
NASA called a team in the 1960 s to investigate the possibility of planting the moon with a small self
Although we still have a long way to go from this goal, Lipsen and Polak have finally demonstrated a computing system that has designed functional machines and has little human intervention
The resulting machine cannot match the fast complexity
Prototype machines designed by human engineers need to be manufactured in practice.
Nevertheless, this is a long wait and necessary step towards the ultimate dream of self
Machine of continuous development