[Docs] Add exploration and training Colab notebooks

[StafaH]

Description

Added two self-contained Colab notebooks in notebooks/ to make Isaac Lab 3.0 easier to explore and learn:

• explore.ipynb: play published locomotion and manipulation policies, switch physics and camera renderers, and run an explicitly untrained cloth demo.
• training.ipynb: create and register an external manager-based ANYmal-D task, check reset/stepping, train state-based PPO, plot learning curves, compare early/final checkpoints, and download the experiment.

Both notebooks include Colab form controls, compact progress displays, expandable logs, and embedded videos. Camera comparisons record the policy's actual RGB input. A README provides launch links and setup guidance. No source packages or project dependencies changed; notebook setup installs the required extras in its runtime.

Review these notebooks directly in Colab before merge:

• Explore
• Training

Type of change

• Documentation update

Release backport

• Backport this pull request to the active release branch after it merges into develop

Validation

• Notebook schema, Python cells, and embedded task-package syntax validated; execution outputs cleared.
• Ruff, formatting, spelling, and repository pre-commit checks passed. The changelog check used a fetched upstream develop reference because the fork's origin/develop was stale.
• Locally validated external task discovery, reset/step, one PPO iteration, cloth training, and a short camera-checkpoint rollout producing 96×96 RGB video.
• Validated progress-helper failure handling, overwritten-video discovery, and all three learning curves against real training output.
• Full hosted-Colab execution and full policy convergence have not been tested.
• The full documentation build is still running locally; no successful result is claimed yet.

Checklist

• I have read and understood the contribution guidelines.
• I have run pre-commit checks with uv run isaaclab -f.
• I have made corresponding documentation changes.
• Notebook-specific syntax, schema, and lint checks pass.
• Targeted runtime checks are documented above.
• Changelog fragments are not applicable: no source packages changed.
• My name already exists in CONTRIBUTORS.md.

[isaaclab-review-bot]

Isaac Lab Review Bot

Reviewed the two standalone Colab workflows and their README integration. The claimed contradiction between “kitless backends” and the OV options is not established by the supplied patch alone, so the candidate finding is not actionable.

• Design and architecture: The notebooks intentionally duplicate their subprocess, progress, and video helpers to remain independently executable. Both isolate simulation commands in the repository’s uv-managed environment and use the existing isaaclab CLI rather than coupling Isaac Lab imports to the Colab kernel.
• API: No repository-owned source API, CLI definition, export, or registry is changed. The training notebook creates an external task package and Gym registration only in the Colab runtime, so there is no public-API migration or changelog obligation evident here.
• Implementation: The setup, task registration, training, checkpoint discovery, video capture, process cleanup, and camera callback paths were traced and are internally consistent. Hosted-Colab execution, full convergence, and external checkpoint or asset availability remain validation risks, but the patch documents those limitations.

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-apps]

Confidence Score: 5/5

The documentation-only PR appears safe to merge, with no actionable correctness or repository-rule violations identified.

Summary

Adds two self-contained Colab tutorials and a notebook landing page:

• explore.ipynb installs Isaac Lab, plays published policies across physics and rendering backends, records policy camera input, and demonstrates untrained cloth behavior.
• training.ipynb generates and registers an external ANYmal-D task, smoke-tests it, trains PPO, plots learning metrics, compares checkpoints, and packages the experiment for download.
• README.md provides Colab launch links, setup guidance, runtime expectations, and troubleshooting notes.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  Colab[Colab GPU runtime] --> Setup[Clone Isaac Lab and sync dependencies]
  Setup --> Explore[Explore notebook]
  Setup --> Training[Training notebook]
  Explore --> Policies[Published policy playback]
  Explore --> Camera[Policy RGB camera recording]
  Explore --> Cloth[Untrained cloth demonstration]
  Training --> Package[Generate and register external task]
  Package --> Smoke[Reset and stepping check]
  Smoke --> PPO[Train PPO policy]
  PPO --> Curves[Plot TensorBoard curves]
  PPO --> Replay[Compare early and final checkpoints]
  Replay --> Archive[Download task, logs, checkpoints, and videos]

Loading

_{Reviews (1) · Last reviewed commit: "Add Isaac Lab exploration and training C..."}