Document warp-frontend play commands in the quick start - #6947
Conversation
The warp environments quick start showed only training commands, so there was no documented way to run inference on a policy trained with --frontend warp. Split the quick start into Training and Playing sections so every train example has a matching play example in both the uv and isaaclab.sh tabs, and note that --video is rejected on the warp path.
The quick start listed only train commands, so there was no documented way to run inference on a policy trained with --frontend warp. Pair each workflow's train command with its play command inside the existing code blocks rather than adding a separate section, so the two are read together. Use the canonical --visualizer newton_gl spelling; "newton" has been a deprecated alias since isaac-sim#6658. Note that --video is rejected on the warp path for both train and play, not just play.
…rp-quickstart-play
The paired train/play code blocks carry the same information, so the introductory paragraph only restated them. Keep the quick start to the commands plus the --video note.
There was a problem hiding this comment.
Isaac Lab Review Bot
Reviewed the documentation-only addition of warp-frontend play commands for both direct and manager-based workflows, along with the new video-recording limitation note. The examples consistently pair with the existing training commands in both quick-start tabs.
- Design and architecture: The additions preserve the existing tab-set and code-block organization. The
--videolimitation is placed alongside the warp quick-start commands where users are most likely to encounter it. - API: No public API or CLI contract is changed. The documented play path uses the existing warp frontend, Newton preset, latest-checkpoint selector, and canonical
newton_glvisualizer spelling. No source package is modified, so the source-package changelog requirement is not triggered. - Implementation: The RST indentation, shell continuations, and note directive are consistent with the surrounding document. The direct and manager-based play examples mirror their corresponding training configurations, and the note clearly identifies that recording requires replay through the torch frontend.
No blocking issues. No inline issue met the actionable-evidence threshold; the assessment above records the review feedback.
Automated review; human maintainers own approval decisions.
Greptile SummaryAdds matching inference commands to the Newton warp-environments quick start and clarifies that video recording requires replay through the torch frontend.
Confidence Score: 5/5The documentation-only change appears safe to merge. The new commands use supported frontend, preset, checkpoint, and visualizer arguments, and no concrete failure remains in the playback or video guidance. Important Files Changed
Reviews (1): Last reviewed commit: "Drop the quick-start prose intro" | Re-trigger Greptile |
The quick start now documents a single execution path, so the tab-set has nothing to switch between. sphinx-design still renders a one-item set as an always-checked radio input plus a label, giving a tab button that cannot be toggled, and emits a stray rubric heading in non-HTML builds. Collapse it to a plain code-block with uv run, matching using-cables.rst.
…rp-quickstart-play
…ay' into jichuanh/docs-warp-quickstart-play
|
@StafaH for review |
|
Backported to |
# Description The Newton warp-environments quick start listed only `train` commands, so there was no documented way to run inference on a policy trained with `--frontend warp`. Each workflow's `train` command is now followed by its matching `play` command inside the same code block, in both the `uv` and `isaaclab.sh` tabs, so the pair is read together. Adds a note that `--video` is rejected on the warp path — for `train` and `play` alike, since the guard lives in the shared `apply_video_recording`. Uses the canonical `--visualizer newton_gl`; `newton` has been a deprecated alias since #6658. Fixes nvbug 6531264. Page: https://isaac-sim.github.io/IsaacLab/develop/source/overview/core-concepts/physical-backends/newton/warp-environments.html#quick-start ## Type of change - Documentation update ## Release backport - [x] <!-- backport-active-release --> Backport this pull request to the active release branch after it merges into `develop` ## Validation Every documented command was run end-to-end on an RTX A6000 against `develop`, with `--num_envs 32 --max_iterations 5` on the train runs to bound them: | Command | Result | | --- | --- | | `train --task Isaac-Cartpole-Direct --frontend warp presets=newton_mjwarp` | `Training time: 1.72 seconds` | | `train --task Isaac-Velocity-Flat-AnymalD --frontend warp presets=newton_mjwarp` | `Training time: 7.96 seconds` | | `play --task Isaac-Cartpole-Direct --frontend warp ... --checkpoint latest --visualizer newton_gl` | loaded `model_4.pt` | | `play --task Isaac-Velocity-Flat-AnymalD --frontend warp ... --checkpoint latest --visualizer newton_gl` | loaded `model_4.pt` | | `play ... --frontend warp --video` | rejected with `ValueError: --video is not supported with --frontend 'warp'` | | `play ... --frontend torch --video` (the note's remedy) | wrote `videos/play/clip_0000.mp4` | `uv run --isolated --extra test -- make -C docs current-docs` (`-W --keep-going`) passes with no warnings, after merging `develop`. No changelog fragment: the change touches only `docs/`. ## Checklist - [x] I have read and understood the [contribution guidelines](https://isaac-sim.github.io/IsaacLab/main/source/refs/contributing.html) - [x] I have run the [`pre-commit` checks](https://pre-commit.com/) with `./isaaclab.sh --format` - [x] I have made corresponding changes to the documentation - [x] My changes generate no new warnings - [x] I have added my name to the `CONTRIBUTORS.md` or my name already exists there (cherry picked from commit 55ac56a)
Description
The Newton warp-environments quick start listed only
traincommands, so there was nodocumented way to run inference on a policy trained with
--frontend warp.Each workflow's
traincommand is now followed by its matchingplaycommand inside thesame code block, in both the
uvandisaaclab.shtabs, so the pair is read together.Adds a note that
--videois rejected on the warp path — fortrainandplayalike, sincethe guard lives in the shared
apply_video_recording.Uses the canonical
--visualizer newton_gl;newtonhas been a deprecated alias since #6658.Fixes nvbug 6531264.
Page: https://isaac-sim.github.io/IsaacLab/develop/source/overview/core-concepts/physical-backends/newton/warp-environments.html#quick-start
Type of change
Release backport
developValidation
Every documented command was run end-to-end on an RTX A6000 against
develop, with--num_envs 32 --max_iterations 5on the train runs to bound them:train --task Isaac-Cartpole-Direct --frontend warp presets=newton_mjwarpTraining time: 1.72 secondstrain --task Isaac-Velocity-Flat-AnymalD --frontend warp presets=newton_mjwarpTraining time: 7.96 secondsplay --task Isaac-Cartpole-Direct --frontend warp ... --checkpoint latest --visualizer newton_glmodel_4.ptplay --task Isaac-Velocity-Flat-AnymalD --frontend warp ... --checkpoint latest --visualizer newton_glmodel_4.ptplay ... --frontend warp --videoValueError: --video is not supported with --frontend 'warp'play ... --frontend torch --video(the note's remedy)videos/play/clip_0000.mp4uv run --isolated --extra test -- make -C docs current-docs(-W --keep-going) passes withno warnings, after merging
develop.No changelog fragment: the change touches only
docs/.Checklist
pre-commitchecks with./isaaclab.sh --formatCONTRIBUTORS.mdor my name already exists there