22년 12월 구글에서 발표한 AI 언어 프로그래밍 CaP.
인간이 일상 언어로 이야기 하면 AI로봇이 맥락을 이해한 후 스스로 그에 맞는 프로그래밍 코드를 작성해 명령을 수행하는 로봇 제어 시스템이다.
ChatGTP와는 달리 로봇 산업에 활용 거능성이 큰 듯 하지만, 잠재력은 그에 못지 않을 것이다.
구글 CaP를 간략히 소개해본다.
ChatGPT is the hot topic of the former market recently.
ChatGPT is an interactive artificial intelligence service released by OpenAI in December 2022.
Google announced 'Bard' as a rival to ChatGPT in February.
In December 2022, before announcing the launch of the Vard, Google has already made a big splash in the robot industry.
It is a self-coding program in which Google uses AI language models to write programming codes on its own and execute commands after understanding the context when a person commands in everyday language.
Developed by Google's robotics engineers, this dedicated AI language program for robot motion control is called Code as Policies (CaP).
This means that robots can generate motion control codes on their own under human instructions and achieve what humans want.
This time, we will look at the characteristics of how CaP can actually operate the robot.
The first feature of CAP is that the general public can easily direct commands in natural language.
In fact, even in relatively easy programming languages, programming code writing is a difficult task to understand specific syntax and available tools.
Because CaP translates natural language into machine language 'in real time', there is no need for separate programming for robot manipulation.
The code that detects objects, the code that operates the actuator that moves the robot's limbs, the code that specifies when the work is completed, and expressions such as "Faster" and "Left" are also replaced with precise numbers to automatically generate the code.
It is also possible to learn and control robots by understanding languages other than English and emoji expressions.
The second is to understand the context of sequential commands and perform instructions accurately.
CaP understands speech or typed sentences, finds the ultimate goal, divides it into stages, and allows the robot to execute using all the technologies it has.
It's like a conversation between a person and a robot in a science fiction movie. This coding model can also understand abstract representations and perform tasks.
CaP said, "I spilled my drink. Can you help me?" allows the robot to interpret the command as "Get a sponge to wipe from the kitchen."
Furthermore, metaphorical expressions such as "Willie Wonka," the founder of a chocolate factory in fairy tales, rather than the explicit word chocolate, can also be used to induce chocolate selection.
It looks like a robot understands and operates language like a human.
The third is that a single robot can perform dozens of new tasks.
In order to teach robots new technologies and tasks, they have to write new codes, which inevitably takes a considerable amount of time.
And, unlike humans, one of the most difficult things in robot motion is to generalize trained behavior.
Programming a robot to play table tennis did not mean that it could play other games such as baseball or tennis.
However, the biggest advantage of the CaP coding model is that the robot does not need to rewrite the code to perform other tasks.
A demonstration video released by Google shows a robot performing dozens of different commands.
A robot can perform a variety of complex and diverse movements, including building blocks, classifying garbage, painting, and erasing.
My Logic is Understandable' My theory is undeniably perfect.' It is a famous line of robots from the movie 'Robot Eye,' which deals with the three principles of robots.
As we saw today, at the current rate of robot evolution, it seems that the era will soon come when this word cannot be ignored.
In the era of general-purpose robots that skillfully handle various tasks as human assistants beyond industrial use, we should pay attention to the development of AI technology that will create the future.
[Picture source : Analytics Instight]
