[Docs] Expand the population-based training guide

[ooctipus]

Summary

• Move the PBT guide from the top-level Features navigation into Overview > Reinforcement Learning, alongside the training guide.
• Explain what PBT does, when it is useful, its benefits, costs, and how it differs from multi-GPU training.
• Add an equal-compute PBT-versus-PPO comparison from the DexPBT reorientation study, including the limitation that it is not an Isaac Lab benchmark.
• Document objective selection, current Isaac Lab selection semantics, shared-workspace requirements, launch steps, evaluation, and troubleshooting.
• Replace the duplicated YAML example with a literal include of the shipping RL-Games configuration.

Validation

• uv run --isolated --extra test -- sphinx-build -W --keep-going -j auto docs /tmp/isaaclab-pbt-docs (full HTML build)
• uv run --isolated --extra test -- sphinx-build -b dummy -W --keep-going -j auto docs /tmp/isaaclab-pbt-docs-dummy (post-rebase)
• SKIP=check-changelog-fragments uv run --frozen isaaclab -f
• Verified all newly added external links return successfully.
• Audited the docs tree: one PBT source page, one owning toctree, and no remaining reference to the former Features path.

Notes

This is documentation-only, so no package changelog fragment is required. The local changelog hook was skipped because it compares against the stale origin/develop branch in the fork and reports unrelated upstream package changes; all other pre-commit hooks passed.

[isaaclab-review-bot]

Isaac Lab Review Bot

This documentation-only PR relocates and expands the population-based training guide, integrates it under Overview > Reinforcement Learning, and replaces the duplicated YAML example with a literal include of the shipping RL-Games configuration.

• Design and architecture: The navigation move is internally consistent: the former Features toctree entry and source page are removed, while the replacement page is owned by the Reinforcement Learning toctree. This avoids duplicate documentation ownership and places PBT alongside related training workflows.
• API: No Python APIs, CLI arguments, or configuration defaults are changed. The reviewed compatibility surface is limited to documentation of existing agent.pbt.* overrides, objective-path semantics, RL-Games support, and the existing multi-GPU documentation link.
• Implementation: The new page covers selection behavior, shared-workspace constraints, launch steps, evaluation, and troubleshooting. The literal-include path is consistent with the relocated page and removes duplicated configuration text. A non-blocking maintenance tradeoff is that :start-at: pbt: has no ending boundary, so later content appended below the PBT block in the source YAML would also appear in the guide.

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]

Greptile Summary

The PR relocates and substantially expands the population-based training guide under the reinforcement-learning overview.

• Explains PBT use cases, compute tradeoffs, selection semantics, objective configuration, and differences from multi-GPU training.
• Adds study context, launch and evaluation guidance, troubleshooting, and references.
• Replaces duplicated YAML with a literal include of the shipping RL-Games configuration.
• Updates the documentation toctrees to reflect the page move.

Confidence Score: 5/5

The documentation-only PR appears safe to merge with no actionable defects identified.

The moved page remains reachable through the documentation hierarchy, its included configuration resolves, no stale references to the former path were found, and the documented PBT behavior aligns with the current implementation.

Important Files Changed

Filename	Overview
docs/index.rst	Removes the former Features toctree entry after relocating the PBT guide.
docs/source/overview/reinforcement-learning/index.rst	Adds the PBT guide to the reinforcement-learning overview navigation.
docs/source/overview/reinforcement-learning/population_based_training.rst	Adds the expanded guide; checked implementation-facing claims and documentation references showed no concrete defect.
docs/source/features/population_based_training.rst	Deletes the superseded guide at its former Features location.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  Root[Documentation index] --> Overview[Overview]
  Overview --> RL[Reinforcement Learning]
  RL --> PBT[Population-Based Training guide]
  PBT --> Config[Shipping RL-Games PBT configuration]
  PBT --> Launch[Population launch workflow]
  PBT --> Evaluate[Checkpoint evaluation]

Loading

_{Reviews (1): Last reviewed commit: "Document population-based training workf..." | Re-trigger Greptile}