Download reference/parallel-environments.txt from Travor278/RoboDojo-Task-Examples: direct link, hf CLI and curl.
- Browser
- Download file 1.43 kB
-
https://huggingface.co/datasets/Travor278/RoboDojo-Task-Examples/resolve/main/reference/parallel-environments.txt
- Command line
-
hf download hf://datasets/Travor278/RoboDojo-Task-Examples/reference/parallel-environments.txt
-
curl -L -o parallel-environments.txt https://huggingface.co/datasets/Travor278/RoboDojo-Task-Examples/resolve/main/reference/parallel-environments.txt
1.43 kB
| RoboDojo supports | |
| multi-environment parallelism | |
| in Isaac Sim. Instead of running one scene at a time, Isaac Sim can instantiate multiple environment copies in the same simulation process, improving evaluation throughput. | |
| Your browser does not support the video tag. | |
| Parallel environments running in Isaac Sim. | |
| Parallel environments are configured through the simulation config selected by the top-level | |
| env_cfg | |
| , for example | |
| env_cfg/sim/sim_config.yml | |
| : | |
| Field | |
| Description | |
| scene.num_envs | |
| Number of environment instances to run in parallel | |
| scene.env_spacing | |
| Distance between environment instances in the Isaac Sim world | |
| Example: | |
| scene | |
| : | |
| num_envs | |
| : | |
| 5 | |
| env_spacing | |
| : | |
| 7 | |
| Increase | |
| scene.num_envs | |
| when you need higher throughput. If you see GPU memory pressure, slower rendering, or unstable simulation behavior, reduce this value and tune it according to your hardware. | |
| Parallel environment count | |
| Section titled “Parallel environment count” | |
| The release default is | |
| num_envs: 1 | |
| , which favors reliability. Increase it for throughput once a task is working on your machine. For randomized cluttered tabletop tasks, the eval client may cap the requested parallelism because these scenes import more objects and usually require more CPU/GPU memory. | |
| Related pages | |
| Section titled “Related pages” | |
| Evaluation Issues | |
| — GPU memory and parallelism troubleshooting when running many randomized environments |