IsaacSim 소개 GPU 기반 로봇 시뮬레이션 툴

IsaacSim은 알고리즘을 검증하고 신경망을 훈련하는 데 사용되는 GPU 기반 로봇 시뮬레이션 도구로, ROS와 통합되어 있습니다.

### NVIDIA IsaacSim 소개: 설치 프로세스 및 개요
* 🤖 엔비디아 아이작심은 엔비디아에서 개발한 GPU 기반 로봇 시뮬레이션 툴로, **사용자가 실제 로봇으로 작업하기 전에 3D 환경에서 로봇을 시뮬레이션**할 수 있습니다.
* 🔧 IsaacSim의 주요 장점 중 하나는 시뮬레이션과 훈련에 GPU를 사용하여 **CPU와 GPU 간의 데이터 전송 시간을 단축**한다는 점입니다.
* 🎮 IsaacSim의 시뮬레이션 환경은 **최신 GPU 기술을 활용**하여 실제와 같은 시뮬레이션을 생성하는 **매우 사실적인 환경**입니다.
* 🖥 IsaacSim을 설치하려면 다양한 NVIDIA 도구를 사용할 수 있는 플랫폼인 **Omniverse 런처**를 다운로드해야 합니다.
* ⚙ 옴니버스 런처를 설치한 후 런처를 통해 **IsaacSim을 다운로드하여 설치**할 수 있습니다.
* 📈 IsaacSim을 실행하는 데 이상적인 GPU 사양은 **GeForce RTX 3080**이며, 최소 요구 사양은 **RTX 2070**입니다.
* 🌐 IsaacSim은 ROS(로봇 운영 체제)와 통합되어 있으며, **화면의 3D 객체 작업을 위해 USD 파일**을 사용합니다.

Hello, from today, I will gradually explain about a robot simulator called IsaacSim. I am also currently studying how to use it for my research and will introduce the studying process to you little by little. So, there may be some mistakes in my explanations, and I would appreciate your understanding. First of all, let me explain the tool we will be studying. When you search on Google or similar platforms, you will find tools related to IsaacSim. As for Hangul Day, I haven't seen much about it, and most likely the materials are in English, so I will focus on that for the explanations. There are excellent tutorials available, and the software is continuously being updated. Therefore, I will briefly introduce what IsaacSim is.

Today, I will explain the process of installation. First of all, our tool called IsaacSim is a robot simulation tool that comes from MBTI. When we talk about this robot simulation tool, as you may already know, let's take a look at Gazebo. Gazebo is a tool that allows us to simulate robots in a 3D simulation environment on a computer by integrating with ROS. Therefore, when we develop and research actual robots, before we proceed to work with the physical robot, we need to verify if our algorithms work well by simulating the robot in a simulation environment. So, Gazebo, PyBullet, CoppeliaSim, etc., there are various types of tools for robot simulation, and among them, IsaacSim is one of the tools to use.

I'm not sure if the first thing that came out on NDB was in 2020. But it's one of the latest developments. One big difference is that while other simulations were done based on the CPU, NVIDIA i30 allows GPU-based simulations. So, when you compare it with the existing simulations, you will notice a significant difference. When you see this, in environments like the 3D gaming environment that we are familiar with, or in virtual environments, it will catch your eye. People often say it looks similar to real life, but if you watch Isaac's video for a moment, I will remove the subtitles. When you see this simulation environment, it shows a very realistic picture. The one above is a simulation, and the one below is real. When you look at it this way, the depiction in the simulation is very high. The top is the simulation, and the bottom is real. This is one of the latest technologies in graphics cards.

It's about rate lessons, utilizing that to create a robot simulation environment that is closer to reality has its advantages. One of the main merits of using the Isaac team, which is a representative team, is that it can use GPUs for optical learning. In the past, when it was based on CPUs, we had to transfer data back and forth between the CPU and GPU to train neural networks for deep learning, which took a lot of time. By simulating within the GPU and training within the GPU, the phenomenon of data bottleneck is reduced. As you can see here, you can create multiple environments, which is called gym. By utilizing this, you can perform simulations and training within the GPU, reducing the time needed for data transfer between the CPU and GPU.

You can run a much larger number of simulations at once compared to simulations. These differences and advantages make it one of the tools that attracts a lot of interest. Also, one of the things that are actually using robot simulations is that once the simulation is done, we need to transfer it directly to some hardware like Raspberry Pi or similar environments, install ROS, and run it. First of all, NVIDIA is a company that is a giant in the hardware industry, so they have products like Nano boards that can use GPUs, making it possible to connect them directly. This can also be seen as an advantage. So, in order to utilize these, I am also studying. Now, let's take a look at the Hello world that appears here for the first time. If you look here, the content here is the tutorial provided by MBTS.

There are a lot of tutorials as I mentioned earlier. Now, in this environment where multiple simulations can be run optically, there are tutorials available. Of course, since it is a robot simulation, it should be integrated with ROS. Among them, there are simulations that are integrated with ROS such as ROS 1 and ROS 2. So, learning about those things like the URDF reporter would be very helpful for you. There may be some limitations due to the language being in English, but we will gradually study and explore those aspects. Originally, in the previous versions, it supported Windows more than now, originally it supported Windows, Mac OS, and Linux. It seems that it was used quite a lot in the past. In fact, for ROS 2, from the 2022.2 version, it seems that Windows version is being supported a lot, so we will provide tutorials for you.

When showing it, I think I can show it well enough on Windows. Well, then, probably, in order to install this, you all need to download something called Omnibus. First, you can download it as an indie visual version, and I think I also used it as an indie visual version, and there wasn't a big problem. I'm not sure exactly how it is compatible with the enterprise version. So, you might test it, or for research purposes, the indie version should be sufficient without any problems. When you download this, you need to fill in things like developer, English name, email address, and then a link to download the service will probably appear. I won't disclose personal information, so I won't put that here. By doing this, when you download it, I will briefly show you the downloaded file.

If you look here, I already have the omnibus lunch version installed, but you will be able to download this. So if you run it, it's currently running. Yes, once you finish installing this, an icon called Omnibus Launcher will appear. I have copied this for your convenience. When you click on this, a window like this will pop up. Let me briefly explain about this window. Probably, initially, the cache may not be installed. I have it installed now, so first, you will probably install the cache, and then various installation processes will follow. Basically, follow the basic steps and then I will explain how to install the main program. First, you probably need to log in like this, so you need to create an account with Naver and log in, and then all the processes.

Once everything is completed, I initially installed this thing called Cache, and maybe I'm not exactly sure, but it seems to make it more convenient to use some apps that are not on the bus. So, I installed this and also installed R even numbers. If you go to the Exchange, you will find these things called Ajax V. In fact, in NBDR, Omnibus is now NVIDIA, originally a company that made GPUs. So, based on GPU technology, you can think of it as a place where various tools that can be used in the metaverse, virtual reality, and so on, on the 3000 platform are gathered together. So, there are tools related to Omnibus now, and besides this, there were also things used in the drive auto vehicle sector, quite a variety of tools. So, now, various tools are available, and even those who create 3D characters use them separately.

Alternatively, there are various tools available, so we will start by using tools related to robots. Since our purpose is robot simulation, we will try using AjaxView among various tools. So, if you click on this, I have the latest version installed now, so when it pops up as Launch, I will choose what was there before. Then, you will see an iPhone-like sound here. So, if you click on Install, it will proceed here, and the installation will take place. As far as I remember, the version I installed, 2022.2, was almost 6-point-something gigabytes, almost 7 gigabytes on Windows. So, you need to download and install that, which will take quite some time. Also, just a quick note from the tutorial, when using Ajax, you need to have an NVIDIA GPU for it to run. So, the basic specifications will be explained briefly later on.

Let me explain. So, when we install this library, you will be able to see it, and another thing called NewClass is a file server used in Isaac Omniverse. Typically, I have also installed servers in Omniverse for services. You will also install these things, and probably related to Isaac Sim, demos or such things will be provided through the local server. Therefore, if you install it along with UPlus, the installation process will be like this. I will show you what I have installed to you, so even if the Next button is not showing, I will check if there is any issue on the page. So, first, if the installation is done like this, by using this launcher, let's try running it once. When you click on Launch, a dialog box will appear, showing the console by default.

There are headless methods and headless methods. Headless means that the GUI is not visible, and you can think of what we have selected now as a version where the GUI is visible. So when you click start, the consultation will be executed like this. This is the part that appears because we have set it to show the console window with checkboxes. Now the loading is completed, so it is loading right now. Probably, it will take some time when you first start it, so now it is loading RTX. Based on the fact that it is loading RTX, you can see that I am zooming in and out using the wheel. If you press the wheel and move, it will move like this. This is the window. For example, let me briefly load it and it may take some time. If you see, it is loading bolts and such. This is the loading part, and I will run the simulation here. Then the arm moves.

We are showing the process of capturing a yacht and entering into a contract as a demo. Actually, if you look at the demo in your tutorials, this is the content to be covered in Advanced Tutorials 23. So, you can actually try out the content, there are explanations about the code, and so on. By following these methods little by little, we will create videos in the tens of millions range to show you. So, in the beginning, you might see some installation processes or something like that in the interactions. It would be nice if I could show you each step one by one, but for now, I haven't done that, so I will explain a bit more while looking at these pictures. This is now something that can be integrated with ROS and such, and basically, in this iXmap, we can use USD files for assignments or when dealing with 3D objects on the screen.

What is a URDF file and what is an SDF file? It means using USD files. So, there are tools to convert the STM file or URDF file to USB before that. Now, I will explain and provide a tutorial. First of all, the ideal state is having a GeForce RTX 3080. Currently, I have a Windows running on a 3090. So, it needs to be at a certain level to run. The best idea is the RTX A6000 series, but you don't have to go that far. The minimum specification is 2070. Therefore, a GeForce RTX series graphics card needs to be installed. I had a gaming laptop with an RTX 3060 before, and even up to 3060, basic tutorials were fine. However, in reality, because GPUs are expensive, you need to have a GPU installed to use it. And there might be some installation steps, let me briefly explain.

I will take a look. The launcher, if you sign up and download it, there are versions for Linux and Windows. So, the Windows version that briefly appeared on the screen earlier can be installed. So, install it, run the launcher, choose and proceed, and it will launch like this, indicating that the installation is complete. And as I showed you earlier, in the selector tool, maybe I also installed it. Depending on the environment, there were several issues, so it might work at once in the beginning, or you might struggle a bit with the installation. It would be good to try installing it once. This is probably based on Docker standards or there will be explanations. So, these things, and this is cloud-based, and Python. Since almost everything is based on Python, after installing Python and setting up the Python environment, we can also debug, you know.

I am explaining the process to you, and I will provide additional explanations when that time comes. This is using Ross OneTwo. If you look at the picture now, there is something called Ross Bridge. There are One Bridge and Two Bridge, depending on the in-app at D7. We can use Ross One and Ross Two. Regarding the setup, if you proceed with the installation like this, install it, and as I showed you in the picture earlier, once you load this Ignition file, you can continue to the next step without any major issues. So, I have briefly introduced basic installation, execution scenes like this to you today. From now on, I will explain how to use it through the actual process or, basically, following the tutorial provided by Isaac. Thank you.

[Isaac Sim - 초급] #1 Isaac Sim 소개 및 설치

동아대학교 기계공학과 
지식-지능형 메카트로닉스 실험실
(Knowledge-Intelligence Mechatronics Laboratory)


KIMe Lab

아이작심 소개: GPU 기반 로봇 시뮬레이터

아이작심은 GPU 기반 시뮬레이션이 가능한 로봇 시뮬레이션 툴로, 가제보와 같은 CPU 기반 시뮬레이션에 비해 더욱 사실적인 3D 환경을 제공합니다.

고퀄리티 그래픽과 딥러닝에 GPU를 사용하여 데이터 전송 시간 및 병목 현상을 줄일 수 있는 장점을 제공하는 최신 개발 제품입니다.

또한 아이작심은 GPU 내에서 시뮬레이션과 훈련을 위한 여러 환경(짐)을 생성할 수 있어 궁극적으로 CPU와 GPU 간 데이터 전송에 필요한 시간을 단축할 수 있습니다.

로봇 시뮬레이션의 장점과 ROS와의 통합

로봇 시뮬레이션은 한 번에 더 많은 수의 시뮬레이션을 실행할 수 있는 이점이 있어 많은 관심을 받고 있습니다.

GPU가 탑재된 나노 보드와 같은 NVIDIA의 제품은 직접 연결이 가능하여 하드웨어적인 이점을 제공합니다.

시뮬레이션은 ROS 1 및 ROS 2와 같은 ROS와 통합되어 있어 ROS 통합을 위한 URDF 리포터와 같은 측면에 대해 학습하는 데 유용합니다.

옴니버스 런치 버전 설치하기: 단계별 가이드

옴니버스 런치 버전은 네이버 계정을 생성하고 로그인한 후 다운로드하여 설치할 수 있습니다.

설치 과정에는 캐시, R 짝수, 로봇 시뮬레이션을 위한 AjaxView 설치가 포함됩니다.

이 프로그램은 메타버스, 가상현실, 3D 캐릭터 제작 등 다양한 도구를 제공하며, AjaxView를 실행하려면 NVIDIA GPU가 필요합니다.

최신 버전의 설치 파일 용량은 약 6~7기가바이트이므로 다운로드 및 설치에 다소 시간이 걸릴 수 있습니다.

아이작 옴니버스의 뉴클래스 라이브러리 설치 프로세스

NewClass 라이브러리는 런처를 사용하여 설치되며, 보이는 GUI 인터페이스로 실행할 수 있습니다.

설치하는 동안 콘솔 창이 표시될 수 있으며, 이 과정에는 RTX 로딩과 시뮬레이션 실행이 포함됩니다.

이 라이브러리의 기능에는 요트 캡처, 계약 체결, 코드 및 데모 탐색을 위한 고급 튜토리얼 제공 등이 포함됩니다.

또한 NewClass 라이브러리는 ROS와의 통합 및 iXmap에서 3D 오브젝트 조작을 위한 USD 파일 사용도 제공합니다.

로봇 공학용 URDF 및 SDF 파일 이해하기: GPU 설치 튜토리얼

URDF 및 SDF 파일은 로봇의 구조와 특성을 설명하기 위해 로봇 공학에서 사용되며, 설치를 위해 GeForce RTX 시리즈 GPU가 필요합니다.

설치 과정에는 런처를 다운로드하고 Windows 버전을 선택해야 하며, 초기에는 몇 가지 어려움을 겪을 수 있습니다.

설치는 Docker 표준을 기반으로 하고 클라우드 기반이며 Python에 의존하며, 설치가 완료되면 로봇 시뮬레이션 및 디버깅에 파일을 사용할 수 있습니다.