机器人能够意识到自己的错误吗 分享到:
Can a Robot Know When It’s Wrong? 2018-08-1064318

双击原文单词查看解释

Robots welding chasis at this car factory are faster and more accurate than humans, and they never need a break.But if these robots make a mistake, they cannot go back and correct it.This is a hard problem because it requires a level of contextual information that we typically don’t see in robot systems.
在这家汽车厂,机器人焊起底盘来比人类更精准、速度也比人类更快,而且机器人不需要休息。但是如果这些机器人犯了错误,它们无法返回纠正。这是一个难题,因为纠正错误需要一定程度的背景信息,而通常情况下机器人的系统中不存在这样的信息。
Henny Admony and her students are trying to teach this robot named Baxter to act as a grocery clerk packing items into a bag.A combination of standard and infrared cameras helps Baxter recognize the objects.For us, the really interesting part is not necessarily the perception of the robot, but how does it incorporate that information into its reasoning system, and what is the underlying reasoning system look like.
Henny Admony和她的学生正在试着教这个名为Baxter的机器当杂货店员,把货物放到包里。标准摄像机和红外摄像机的组合帮助Baxter识别物体。对我们而言,真正有趣的部分并不一定是机器人的感知能力,而是机器人接收信息、并把信息纳入自己的推理系统的方式,以及其内在推理系统的运作方式。
But why don’t program errs incorporate all possible errors and potential fixes into the algorithm that drives the robot.It’s not particularly efficient because they might miss an error in this scenario and when they move to a new situation,when the robots manipulating some other object now they need to redefine all those errors again.So right now, researchers are trying to figure out how the robot takes in and interprets information in a way that lets it do the kind of reasoning they want it to do.It’s pretty difficult. It sounds really abstract. We do it as humans very very naturally, but we need to build these mathematical models so that robots can be able to do it, too.The four-year project is being done in collaboration with Brigham Young University, Tufts University and the University of Massachusetts.
但是为什么不用编程把所有可能的错误和潜在解决方式纳入驱动机器人运作的算法内呢?这种方法不是很有效,因为可能会漏了这种情况下的某个错误,而机器人进入到新的情境中,或者他们的操作对象发生变化时,它们需要重新定义所有的这些错误。所以现在,研究人员正在试着探究出机器人接收、解释信息的方式,以让机器人能照着他们的方式推理。这个很困难,听起来很抽象。我们是人,推理起来十分自然,但是,想要让机器人也能推理,我们就需要建立这些数学模型。这项为期四年的项目正由杨百翰大学、塔夫茨大学和马萨诸塞大学合作完成。
As for robots that can fix their own errors and recover from mistakes, Admony says that technology may still be 10 to 15 years away.
至于能够修复自己错误并从错误中恢复的机器人,技术上还需要花费10年到15年的时间。
George Putic , VOA news, Pittsburgh.
美国之音新闻,George Putic华盛顿报道。

点击加载更多