scene-gaussian-map-alignment
Skill7 starsChanged 53 days agoMITView on GitHub
Align scene Gaussian/splat, USD/mesh, and robot map assets into an honest digital-twin evidence workflow. Use when a new Gaussian scene arrives, when B1/Map12-style assets need to be connected, when map anchors are being projected into a 3D scene, or when an agent must decide whether an alignment is candidate, verified, runtime-proven, or planner-backed.
In this repository8 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- eval-evolutionSkill
- eval-harnessSkill
- household-worldSkill
- raw-fpv-visual-labelerSkill
- report-performance-analysisSkill
- visual-result-showcaseSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. About 0 are common, 10 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 8
Use this when a bundle combines a 3D Gaussian/splat/PLY/USD/OBJ scene, a robot map such as Nav2 YAML plus occupancy PGM, semantic anchors …
- Distinctive: line 15
Keep stable, repeatable work in scripts.
- Distinctive: line 22
Do not hide scene-specific assumptions in a script default.
- Distinctive: line 28
Use these labels consistently:
- Distinctive: line 40
Never skip tiers in wording.
- Distinctive: line 50
Run the deterministic tools that apply to the available assets:
- Distinctive: line 93
Write the lightweight alignment manifest.
- Distinctive: line 111
Do not claim Gaussian fusion unless the renderer consumed the Gaussian/splat asset.
- Very specific: line 113
Do not claim semantic_anchors_are_usd_truth=true without segmentation, object manifest, or anchor correspondences that bind map anchors to …
- Distinctive: line 116
Do not claim manipulation support without object/receptacle binding plus a pick/place proof.
- Very specific: line 121
If Map 12 semantics are used only as navigation-memory anchors, call them robot_map_12_navigation_memory_overlay or an equivalent overlay …
- Distinctive: line 127
When handing off results, include:
- Distinctive: line 140
After changing this skill, related scripts, or map/Isaac report contracts, run:
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Scene Gaussian Map Alignment
Use this when a bundle combines a 3D Gaussian/splat/PLY/USD/OBJ scene, a robot
map such as Nav2 YAML plus occupancy PGM, semantic anchors from
navigation_memory.json, or an Isaac/operator-console report that must label
evidence honestly.
Boundary
Keep stable, repeatable work in scripts. The skill owns the judgment that changes from scene to scene:
- Stable scripts parse headers, bounds, map metadata, semantic-memory JSON, transforms, smoke artifacts, images, and HTML reports.
- The skill decides the evidence tier, asks for missing assets, chooses which anchors are credible, names blockers, and prevents overclaiming.
- Do not hide scene-specific assumptions in a script default. If an assumption will vary with the next Gaussian scene, write it in the skill/report as an explicit decision.
Evidence Tiers
Use these labels consistently:
blocked: geometry, map files, semantic anchors, or coordinate evidence are missing.candidate: bbox fit, scale/translate, manual placement, or another heuristic alignment exists.verified: named physical/semantic anchors match across map and scene with residuals recorded.runtime_proven: Isaac or robot-view smoke renders/navigates through candidate waypoints and writes view evidence.planner_backed: a real planner/Nav2-equivalent path proof exists.
Never skip tiers in wording. A runtime smoke can prove that rendered robot views exist at candidate poses; it does not by itself prove Nav2 planner parity.
Workflow
- Inventory every asset before aligning: Gaussian/splat/PLY files and whether they are rendered or only inspected, USD/OBJ/mesh world bounds, Nav2 YAML, occupancy grid, semantic memory, map-bundle context, anchor ids, and any segmentation/object manifest/correspondence/calibration evidence.
Run the deterministic tools that apply to the available assets:
python -m roboclaws.household.agibot_map_bundle \ --source-map-dir <map-root> \ --context-json <context-json> \ --output-dir assets/maps/<bundle-name> .venv-isaaclab/bin/python -m roboclaws.backends.isaaclab.b1_readiness \ --b1-root <scene-root> \ --map12-root <map-root> \ --output output/<run>/readiness.json .venv-isaaclab/bin/python -m roboclaws.backends.isaaclab.b1_navigation_smoke \ --b1-root <scene-root> \ --map12-root <map-root> \ --output-dir output/<run> \ --accept-nvidia-eula .venv-isaaclab/bin/python -m roboclaws.backends.isaaclab.b1_readiness \ --b1-root <scene-root> \ --map12-root <map-root> \ --navigation-artifact output/<run>/navigation_smoke.json \ --require-navigation-success \ --output output/<run>/readiness_with_navigation.json python -m roboclaws.backends.isaaclab.b1_navigation_report \ --run-dir output/<run>- Classify what the run actually proved: bbox fit is
candidate; matched anchors with residuals areverified; rendered robot views at candidate poses areruntime_proven; planner path evidence isplanner_backed. Summarize the evidence without changing the artifacts:
python skills/scene-gaussian-map-alignment/scripts/summarize_alignment_evidence.py \ --readiness-artifact output/<run>/readiness_with_navigation.json \ --navigation-artifact output/<run>/navigation_smoke.json \ --output output/<run>/alignment_evidence_summary.jsonWrite the lightweight alignment manifest. This is the fusion contract for future runs; it is not a fused USD/Gaussian scene:
python skills/scene-gaussian-map-alignment/scripts/summarize_alignment_evidence.py \ manifest \ --readiness-artifact output/<run>/readiness_with_navigation.json \ --navigation-artifact output/<run>/navigation_smoke.json \ --evidence-summary output/<run>/alignment_evidence_summary.json \ --map-bundle assets/maps/<bundle-name> \ --output output/<run>/alignment_manifest.json- Report the open blockers and next promotion step. Prefer one precise blocker over broad language like "alignment done".
Honest Labels
- Do not claim Gaussian fusion unless the renderer consumed the Gaussian/splat asset; header/bounds inspection is only inventory evidence.
- Do not claim
semantic_anchors_are_usd_truth=truewithout segmentation, object manifest, or anchor correspondences that bind map anchors to USD/scene objects. - Do not claim manipulation support without object/receptacle binding plus a pick/place proof.
- A "verify image" is a rendered camera view from a candidate pose. It helps inspect gross placement and visibility, but it is not ground-truth alignment, semantic binding, or planner proof.
- If Map 12 semantics are used only as navigation-memory anchors, call them
robot_map_12_navigation_memory_overlayor an equivalent overlay source, not USD truth.
Output
When handing off results, include:
- alignment tier and transform source;
- whether Gaussian assets were rendered or only inspected;
- semantic source and semantic/USD binding status;
- navigation evidence status and whether it is planner-backed;
- artifact paths such as
readiness.json,navigation_smoke.json,readiness_with_navigation.json,alignment_evidence_summary.json,alignment_manifest.json,report.html, and any map bundle; - the exact next step needed to promote the evidence tier.
Acceptance
After changing this skill, related scripts, or map/Isaac report contracts, run:
./scripts/dev/run_pytest_standalone.sh \
tests/contract/maps/test_b1_map12_digital_twin_readiness.py \
tests/contract/maps/test_b1_map12_navigation_report.py \
tests/contract/maps/test_agibot_map_bundle_export.py \
tests/contract/skills/test_scene_gaussian_map_alignment_skill.py \
tests/contract/skills/test_skill_manifests.py \
-qDoes it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

