Gene algorithm makes computer design self-evolution
Through evolutionary magical power, nature created human; today, European and American scientists are trying to be in computer software tools
Establish a certain evolutionary mechanism to allow the computer to consciously experience the evolutionary process, and the animation and engineering designers have become a relaxed command.
And a happy audience on the computer screen, an animated character explores one step, trying to walk. But it fell immediately.
It's helpless. After a while, it appeared on the left side of the screen, just a few steps like a baby, and fell. again
After returning to the screen, it lifted his knees and walked forward to six or seven steps before, and then fell again ... it once again.
I tried a second time, I finally walked like a real person.
This short film will not win Oscar Best Cartoon Award, but computer software manufacturing this effect is called in history.
The miracle, because this animated person is to learn to walk. This intelligent technology is attributed to some smart programs, this
Some procedures imitated the process of screening success genes in nature.
"Evolution" is a random strain
Scientists call this programming as "gene algorithm": given problems, first randomly identify some potential solutions and carry out
Preferably, the "gene" of this batch of success is accompanied by some of the next generation of "variation" to them. this process
Continue to repeat until the problem of evolving problems.
The gene algorithm was the earliest John Holland, the University of Michigan, is proposed in the 1960s. Now it is more and more
Multi-use to actual production, such as a refrigerator with a higher refrigeration efficiency.
The gene algorithm makes the computer to do some things that look complicated, such as thinking. The animator who learned to walk
For example, a computer does not rely on human guidance and make an example of deciding. Therefore, it is the fastest from this technology.
Will be the mass entertainment industry. Hollywood's large and action games are filled with computer manufacturing. In order to make it
They are more realistic, they must let them learn to walk.
Not long ago, the manufacturer has to do this, but have to gradually design how to move each arm, or move the real person
The three-dimensional image is used to a virtual animator. This requires a massive calculation work, and even worse, this
The method does not have flexible variability. If the animator first desigrates a cartman from the mountain, then he changed the master.
It is necessary to let the animator skip a rock, he must design the action from the beginning.
"Evolution" can save mental strength
Tolstten Riro is a researcher in Oxford University, and later made an animation. He decided to learn from nature
The power of the medium evolution is to solve this problem, making a more realistic digital animator. He explained: "We
A character frame is created first, it has weight, the section. Then we give it a virtual muscle and control these muscles.
Nervous network of meat. The problem is, how do you make these networks work as you ask? If you are just randomly
Building a network, it will convey a complex signal to the muscles, which is often unable to walk the animator, it is just a critical convulsion
. "In this case," muscle "can be moved, and they can also be connected to" nerve central system ", but the animator is still
I don't know how to walk.
A cartoons must walk naturally as living as a live person, and more than 700 parameters need to be optimized. Ray said: "Once
You carefully watch this animation system, you know that you can't solve this problem at all, because
They are too complicated. Therefore, we need to utilize the power of evolution. "
"Evolution" 20-generation learning will walk
Ray and his team have created a gene algorithm to explore whether it can accurately regulate the animation role control system. base
It is relatively simple because the elements of the algorithm are relatively simple. It includes some "organisms" containing different "genes"; determine these
Genetic combination and variation of laws; adaptation functions, that is, to evaluate which of all generations are the best - in animation, the evaluation criteria for this function are "from the origin to the distance from the origin."
Rail uses this algorithm to create 100 animation characters, each of which is controlled by randomly combined nerve central. immediately
Let them try to walk. As expected, the first generation cannot achieve walking function. But there are several performances to be a little better.
Some, after all, take a step forward. Based on the standard of adaptation, they become winners in the first round. gene
The algorithm software copies 20 copies according to their nerve center, and has a little variation to it, and adds 80 new.
For the participants, carry out the next generation of walking experiments.
"Evolution" requires a few worsens
Like organic life, gene algorithm is also spacious and neoquiry. Some algorithms and "mating",
Re-combined genes. Some of the programs that are just cloned and the results of the mutation.
The application of a gene algorithm is inevitably brings a miracle. Ray's cartman quickly improved his ability to walk.
Solve. However, they don't always go very well. Some are not going at all but go forward or crawling. These ones
The role is just the mechanical execution of the game rules, so Rally has to change the rules of the game: "We have to add a few
Working - success is not just based on how much walking distance, but is not less than one according to the animated center of gravity
The distance passed by the point is judged. "
In the end, Rayl only used 20-generation movies, which was the optimization process for a few minutes. Group made
The video shows a few generations of evolutionary processes: from the first generation that will not walk to the twentieth generation success.
Self-evolutionary cartman is just an example of many gene algorithms. Look at the recording of its evolutionary history, people are not
Lenovo, this virtual evolutionary process and millions of years ago, our ancestors learned to walk in the African grassland.
Asphap. Those animated characters can walk, not which high person is assembled, but the evolutionary process makes them
Can establish and select a suitable movement and muscle control mode.
"Evolution" algorithm is unlimited
Rail's animation gene algorithm did not make the computer self-awareness, but this algorithm did make computer more creative.
It allows the animator to surpass the intentional control of the human engineer, and strive to self-improvement. Now, Rail design
The generator algorithm has been incorporated into an animated design software called "endorphin".
Applying gene algorithm technology to actual is more than Ray and his group. A development of "Creative" in California, California
Ming Lab, Bill Gros and his inventions are also developing solar equipment using gene algorithms. Gross phase
The letter, the genetic algorithm can trigger a revolution in engineering design, no longer only applying a software as an expression design
The auxiliary tools of ideas, but the gene algorithm will be able to design independently.
People only need to define organisms, genes, and adaptable functions, and the rest is to let software completed the hardships of that series.
Calculate. For example, you have to do a table, you don't have to draw a table, you just need to give your condition: How high, more wide, edge
When you wait, then you tell these designs software, which gives the best answer at the lowest cost. This
It is an ideal state of engineering design!
Excerpt from [International Herald]
