May 22, 2013 07:48am ETSHow Evolution May Help Build Better Robots Smart Machines This virtual robot evolved to move as quickly as possible using four types of tissue: soft support tissue, bone and two types of muscle. While a living creature has cells, this robot has voxels (or 3-D pixels) colored according to the type of tissue. Credit: Creative Machines Lab View full size image

NEW YORK — In the real world, animals have evolved the ability to get from point A to B by galloping, crawling and jumping. Now, robots in the virtual world have accomplished something similar.

In new work, researchers have simulated evolution using virtual robots and watched them develop locomotion strategies of their own.

In robot-creating simulations, researchers started with random assortments of four types of tissues — including two kinds of muscle, soft support tissue and bone. The simulations favored the tissue configurations that traveled the fastest from point A to point B. Then the team allowed the mathematical simulation to run its course over 1,000 generations of robots.

"We see really cool stuff as a result of that, without any interaction from me or anyone else, just this process unfolding itself," Nick Cheney, a member of the research team and a doctoral student at Cornell University, told an audience of reporters Tuesday (May 21) here in midtown Manhattan.

The team dubbed the categories of successful robot design that emerged as the L-Walker, the Incher, the Push-Pull, the Jitter, the Jumper and the Wings. [Super-Intelligent Machines: 7 Robotic Futures]

One of the virtual robots demonstrates the strategy for locomotion it has developed through simulated evolution. Credit: Creative Machines LabView full size image

"I would never come up with anything that looks remotely like that," Cheney said, referring to one of these virtual robots. The bots consist of cubes known as voxels (three-dimensional pixels), which display bright colors signifying different types of tissue.

In these simulations, the virtual robots accomplished something highly unusual for robots: They adapted.

Most robots currently in use in the real world are precisely engineered to work in highly constrained environments, such as manufacturing floors, with their every action hand designed and coded by engineers. As a result, these machines cannot adapt to unfamiliar surroundings.

Unlike human engineers, however, nature is a master at creating creatures that can adapt to and interact with their surroundings. This happens through natural selection, the process by which certain traits give organisms a better chance to survive and thus produce more offspring. Nature thus "selects" these traits to persist in future generations. Cheney and colleagues are striving for a similar process in robotics.

Although the creatures he and colleagues created do not currently exist in the real world, they could be created with 3D printing.

"The truth of the matter is we can print almost anything, any design," he said, noting researchers recently made an artificial ear with living cells using a 3D printer.

In creating the virtual, soft-bodied robots, the team intentionally avoided the traditional robotics' design approach, Cheney said.

"We wanted to be true to nature and introduce muscles and bones and tissues," he said.

Most of the random assortments of tissues that served as a starting point were "pretty bad," he said. "Every once in a while, you get lucky and one is slightly better. Those reproduce more … Over time, you get some pretty amazing things."

In real life, a molecule called DNA (deoxyribonucleic acid) encodes the instruction set to create a living organism; analogously, these virtual robots were created using what is known as a compositional pattern-producing network, or a network of mathematical functions, Cheney said.

Many of the strategies that emerged among the soft-bodied robots mimicked those of animals, such as a galloping horse or a crawling inchworm.

The research team included Cheney, colleagues Robert MacCurdy and Hod Lipson of Cornell's Creative Machines Lab, and Jeff Clune of University of Wyoming's Evolving AI Lab. The research is scheduled for presentation at the Genetic and Evolutionary Computation Conference in Amsterdam in July.

展示进化可能有助于构建更好的机器人

SHow进化可能有助于建立更好的机器人智能Machines 这个虚拟的机器人发展到那样迅速行动尽可能使用四种类型的组织：支持软组织，骨和两种类型的肌肉。而一个生物具有细胞中，这种机器人具有体素（或三维像素）根据组织的类型着色。图片来源：创意机械实验室查看原图

NEW YORK＆mdash;在现实世界中，动物进化从A点奔腾，爬行和跳跃到B的能力。现在，机器人在虚拟世界中已经完成了类似的东西.

在新的工作中，研究人员模拟进化使用虚拟机器人，看着他们发展自己的运动策略。 ＆NBSP;

在机器人创造模拟，研究人员从四种类型的组织和mdash随机花色品种;包括二种肌肉，软支持组织和骨的。仿真赞成组织配置，旅行速度最快的从A点到B点。那么球队允许的数学模拟超过1000一代机器人自生自灭.

＆QUOT;我们看到很酷的东西，那结果，未经我或任何其他人的互动，只是这个过程中展现自己，＆QUOT;尼克·切尼，研究团队和博士生在康奈尔大学的成员，告诉观众的记者周二（5月21日）在这里曼哈顿中城.

队被称为是成为了L-沃克成功的机器人设计的类别，该Incher，推挽，抖动，跳线和翅膀。 [超级智能机：7机器人期货Super-Intelligent Machines: 7 Robotic Futures图片来源：创意机器的LabView原图

[我永远不会拿出任何远程这样看起来，＆QUOT;切尼说，指的是这些虚拟的机器人之一。该机器人由被称为体素（三维像素），这显示鲜艳的颜色标志着不同类型的组织.

在这些模拟多维数据集，虚拟机器人完成一些非常少见的机器人：他们在适应.

大多数的机器人目前在现实世界中使用的精确设计高度受限的环境，如制造工厂工作，他们的每一个动作的手设计，由工程师编码。这样一来，这些机器无法适应陌生的环境.

不像人类的工程师，然而，大自然是主创造的生物，能够适应并与周围环境互动。出现这种情况通过自然选择的过程，其中某些特性给生物一个更好的生存机会，从而产生更多的后代。因此，自然与QUOT;选择＆QUOT;这些特质坚持后代。切尼和他的同事正在努力争取一个类似的过程在机器人.

虽然他和同事们创造了当前不存在于现实世界中的生物，它们可以用3D打印.

u0026 QUOT创建，事情的真相是，我们可以打印几乎所有的东西，任何设计，与QUOT;他指出，研究人员最近提出一种人造耳朵使用3D打印机.

在创建虚拟，软体机器人的活细胞，球队有意避免了传统机器人??“的设计方式，切尼说.

u0026 QUOT;我们希望是真实自然，引进肌肉和骨骼和组织，与QUOT;他说.

大多数组织的服务为出发点的随机花色品种都是＆QUOT;非常糟糕，和QUOT;他说。 ＆QUOT;每过一段时间，你很幸运，一个稍微好一点。这些繁殖更多＆hellip;随着时间的推移，你会得到一些非常了不起的事情＆QUOT;

在现实生活中，一个被称为DNA（脱氧核糖核酸）分子编码的指令集来创建一个活的有机体;类似地，使用了被称为一个组成图案生产网络，或数学函数的网络中创建这些虚拟机器人，切尼说.

许多该软躯体机器人之间出现的策略的模仿那些动物，如走马或爬行蠕动.

该研究团队包括切尼，同事罗伯特MacCurdy和康奈尔大学的创新机器实验室利普森和怀俄明不断发展的人工智能实验室大学的杰夫Clune。该研究定于演示的遗传与进化计算会议在阿姆斯特丹月.