📡 Embodied AI Radar具身智能雷达

Topics主题 › Sim-to-Real & Simulators仿真与 Sim-to-Real

Paper论文 2021 · NVIDIA

Isaac Gym: High Performance GPU-Based Physics Simulation For Robot Learning

Keeps physics and policy on the GPU end-to-end, enabling thousands of parallel environments on one card.物理仿真与策略全程在 GPU 上运行,单卡即可并行数千个环境。

BibTeX
@misc{makoviychuk2021isaac,
  title = {{Isaac Gym: High Performance GPU-Based Physics Simulation For Robot Learning}},
  author = {Viktor Makoviychuk and Lukasz Wawrzyniak and Yunrong Guo and Michelle Lu and Kier Storey and Miles Macklin and David Hoeller and Nikita Rudin and Arthur Allshire and Ankur Handa and Gavriel State},
  year = {2021},
  eprint = {2108.10470},
  archivePrefix = {arXiv},
  primaryClass = {cs.RO},
  url = {https://arxiv.org/abs/2108.10470},
}

Authors (from arXiv): 作者(来自 arXiv):Viktor Makoviychuk, Lukasz Wawrzyniak, Yunrong Guo, Michelle Lu, Kier Storey, Miles Macklin, David Hoeller, Nikita Rudin et al.

More from: 同机构更多条目:NVIDIA英伟达

More in Sim-to-Real & Simulators更多仿真与 Sim-to-Real条目

All Sim-to-Real & Simulators entries →查看全部仿真与 Sim-to-Real条目 →