Isaac Gym RL Benchmark Environments Installation and Analysis

แชร์
ฝัง
  • เผยแพร่เมื่อ 30 ก.ค. 2024
  • Installation and Argument Analysis of the Isaac Gym Benchmark Environments
    00:00 Intro
    00:24 Github access
    10:39 Creating an environment
    21:11 Running the benchmarks
    28:22 Loading trained models, checkpoints
    39:49 Configuration and command line arguments
    41:27 Tasks
    43:05 Domain Randomization
    44:54 Reproducibility and Determinism
    46:42 Multi-GPU Training
    49:39 Population Based Training
    51:25 WandB support
    53:23 Capture videos
    57:42 Capture videos during training
    01:03:04 Pre-commit
    01:03:54 Troubleshooting
    01:04:28 Citing
    01:05:23 Ending
  • วิทยาศาสตร์และเทคโนโลยี

ความคิดเห็น • 7

  • @robotjang01
    @robotjang01 11 หลายเดือนก่อน +1

    혼자 독학으로 isaac gym 시뮬레이션 환경을 구축중이었는데 덕분에 많은 도움이 됩니다~ 감사합니다!!

  • @kyw0615
    @kyw0615 21 วันที่ผ่านมา

    multi gpu 쓸 때 질문이 있는데요. 대부분의 메인보드가 두 개의 gpu를 사용하면 pcie 배속을 하나는 16x랑 하나는 4x로 지원을 하더라고요. 다른 경우는 8x 8x정도가 있고요. 그리고 워크스테이션용 cpu가 아닌 이상 16x 16x를 다 받아줄 수도 없잖아요. 말이 좀 길었는데요. 궁금한 건 isaac gym의 경우 pcie 배속에 성능이 얼마나 영향을 받나요? 이번에 큰 맘 먹고 ai용 pc 구매할 예정이라 질문해봤습니다. 아 그리고 영상 잘 봤습니다. 유익하네요.

    • @tweak-cd6ss
      @tweak-cd6ss  20 วันที่ผ่านมา

      pcie 배속을 비교/변경하며 isaacgym에서 multi-gpu를 이용한 벤치마크 실험을 해본적은 없습니다. 제가 사용하는 PC의 메인보드는 두 개 GPU활용 시 x8/x8로 제한되는 모델인데요, 질문에 정확히 답변이 되는 실험은 아니겠으나 Humanoid 태스크를 1000epoch 학습 시킴에 있어서 single-gpu는 약 523초, multi-gpu(RTX6000 ada 2개)는 약 552초 가량 소요되어 overhead로 인하여 multi-gpu가 학습에 더 많은 시간이 소요되었습니다(state기반, not image). pcie 배속에 따른 성능 측정을 위해서는 x4레인으로 강제로 다운 시켜서 multi-gpu 학습 후 비교하면 될 것 같은데, 다른 PC는 사양이 달라서 정확한 비교가 안될 것 같고, 지금 PC는 재부팅이 불가능한 상황이라 더 이상의 실험은 어렵네요...
      개인적인 생각으로는 isaacgym과 연동하여 학습할 모델이 큰 이미지를 다루고 거대한 모델을 다루는 경우, x8/x8, multi-gpu 조합이 유리해 보이고, 그게 아니라면 x4/x16-single-gpu로 셋팅하여 first gpu는 display/evaluation용, second gpu는 x16 slot에 물려서 학습 전용으로 구성하는 방법도 좋아 보입니다. 아무래도 x4 x16간에 레인 차에 따른 multi-gpu overhead 및 속도 저하가 예상되기 때문입니다. 만약 저라면 그냥 x8/x8로 갈 것 같습니다~

  • @isaacbalbuca1070
    @isaacbalbuca1070 11 หลายเดือนก่อน

    Hello,
    Inside the Isaac Gym folder, there is a (sh) file, could you please provide me with information about the purpose and application of this environment?
    I'm interested in using isaacgym to track the trajectory of a robotic arm using Deep Reinforcement Learning. I also noticed the presence of "orbit," which interacts with IsaacSim and IsaacGym. Could you kindly guide me on how to get started with these tools?
    Thank you!

    • @tweak-cd6ss
      @tweak-cd6ss  11 หลายเดือนก่อน

      Are you talking about the "isaacgymenvs/pbt/launcher/slurm/sbatch_timeout.sh" file? This file appears to be a shell script used for performing slurm scheduling when you do the population based training. Please refer to the following links
      slurm.schedmd.com/download.html
      www.hpc.kaust.edu.sa/tips/use-%E2%80%9Ctimeout%E2%80%9D-command-automatically-restart-jobs
      Describing Orbit in a comment would be quite lengthy. You can find detailed information and instructions at the following link.
      docs.omniverse.nvidia.com/isaacsim/latest/ext_omni_isaac_orbit.html
      isaac-orbit.github.io/

    • @isaacbalbuca1070
      @isaacbalbuca1070 11 หลายเดือนก่อน

      I mean the "rlgpu" file, which is the specific function of this anaconda environment.@@tweak-cd6ss

    • @tweak-cd6ss
      @tweak-cd6ss  11 หลายเดือนก่อน +1

      That file is just a simple shell script to automate the Isaac Gym installation process based on anaconda environment, as described in "Installation" of user guide.