ContentVault

Content packs for Lucky Robots. Each pack contains a metadata.yaml, a thumbnail.png, and a .zip with the pack contents.

Clone

git clone https://huggingface.co/luckyrobots/ContentVault

Content Packs

Pack	Description
Bunglalow	Bungalow environment
Go2VelocityTracking	Unitree Go2 velocity tracking
Loft	Loft environment
MujocoExample	MuJoCo simulation example
Office	Office environment
Oscillator	Oscillator environment
Panda	Franka Panda robot
Piper	Piper robot
PiperBlockStacking	Piper block stacking task
PiperPickPlace	Piper pick and place task
PiperUnscrewCap	Piper cap unscrewing task
PodGallery	Editable native Pod gallery with materials, articulated exhibits and elastic cables (2026.3+)
RobotCafe	Path-traced cafe where a Unitree G1 serves coffee to the tables (2026.3+)
SO100PickAndPlace	SO-100 pick and place task
Skies-Vol-1	Sky HDRIs volume 1
UnitreeG1	Unitree G1 humanoid
UnitreeGo2	Unitree Go2 quadruped
UnitreeH1	Unitree H1 humanoid model (developer pack; no controller)
UnitreeH2	Standard Unitree H2 humanoid (developer pack; fixed hands, no controller)
UnitreeH2PlusSharpa	Unitree H2 Plus with articulated Sharpa hands (developer pack; no controller)
Welcome	Welcome / intro pack

Pack engine versions

Project and robot packs that require a legacy-to-Pod migration have separate archives for engines through 2026.2 and for 2026.3 onward. The newer folder and ZIP names append Pod to the original pack ID, while keeping the same display name and packs/devpacks membership. For example, CupCleanerPod/CupCleanerPod.zip contains the Pod version of CupCleaner. Welcome and the existing WelcomePod already form this pair; both are indexed under packs.

The public robot packs follow the same pairing:

Legacy pack (before 2026.3)	Pod pack (2026.3 onward)
OpenArmPedestal	OpenArmPedestalPod
Panda	PandaPod
Piper	PiperPod
UnitreeG1ThreeFingers	UnitreeG1ThreeFingersPod
UnitreeG1FiveFingers	UnitreeG1FiveFingersPod
UnitreeGo2	UnitreeGo2Pod
SO100	SO100Pod

Robot Pod archives replace .hmograph files with native .hpodgraph motion graphs. Meshes, policy models, sensor actions and thumbnails are retained. The Go2 and G1 three-finger models use the policy-aligned motor gains already shipped in ApiTestsPod, and G1 policy descriptors reference the adjacent model_config.json. The two G1 Pod packs use distinct C# namespaces and explicit script actions, so they can coexist in one project. Embedded scenes retain their existing script class identities; policy behavior is unchanged. SO100 reuses the native graph from SO100PickAndPlacePod, including its event flow and rig-specific IK settings. Archive roots and action path prefixes keep the original robot ID; the installer resolves them beneath the selected pack's folder. Both variants retain the same display name.

The migrated projects carry the same base robot files as the standalone Pod packs. Scene-specific arm controllers are named MG_Panda_Scenario or MG_Piper_Scenario and keep their existing asset handles and tool-point conventions. SimplePick retains its calibrated Piper gripper gains, and RobotCafe retains its grasping-controller gains and finger contact settings; these are task-specific adaptations. Extra palm sites and scene wrappers are also retained. WelcomePod's shared authored files match the engine's bundled Welcome, including analytic marble colliders, free-rolling contacts and the aligned upper join with a gentle downhill grade. The archive additionally includes the separate generated-layout experiment.

EngineVersion.Max is exclusive: the original packs use Max: "2026.3" to include every 2026.2 release. The new packs use Min: "2026.3". Both use Behaviour: Hide, so incompatible variants disappear instead of becoming disabled cards. Existing lower bounds on the original packs are preserved.

The developer-only AnymalC, Nero and UnitreeGo2Lidar legacy archives also stop at 2026.3 because they still contain .hmograph files.

The new archives use Pod motion and animation graphs, retaining asset handles, scene inputs, clip bindings and script controls. Archives retained for legacy engines are unchanged.

SimplePick already uses Pod and remains one pack with Min: "2026.3" and Behaviour: Hide. BlankProject contains no graphs, remains one pack without an engine-version restriction, and continues to serve the editor's starter cards.

CDN

All content packs are served globally via Cloudflare CDN:

https://contentvault.luckyrobots.com/{PackName}/{filename}

For example:

https://contentvault.luckyrobots.com/Loft/metadata.yaml
https://contentvault.luckyrobots.com/Loft/Loft.zip

How sync works

A Cloudflare Worker (worker/) listens for webhooks from HuggingFace. When a push is made to this repo, the Worker:

1. Parses the commit diff to find changed files
2. Streams each file directly to Cloudflare R2 (no buffering, handles large zip files)
3. Deletes removed files from R2
4. R2 serves the files through Cloudflare's CDN with a 30-day cache at 300+ edge locations worldwide

Worker setup

cd worker
npm install

Secrets (set via wrangler)

npx wrangler secret put HF_WEBHOOK_SECRET
npx wrangler secret put HF_TOKEN

Deploy

npx wrangler deploy

Setup

# Install huggingface_hub with fast upload backend
pip install -U "huggingface_hub[hf_xet]"

# Login (requires write-access token from https://huggingface.co/settings/tokens)
hf auth login

---

CLI Commands

Upload entire folder

hf upload luckyrobots/ContentVault <local-folder> . --commit-message "Your message"

Upload a single pack/subfolder

hf upload luckyrobots/ContentVault <local-folder>/<pack-name> <pack-name> --commit-message "Add <pack-name>"

Upload a single file

hf upload luckyrobots/ContentVault <local-file> <path-in-repo> --commit-message "Add file"

Upload large folder (resumable, multi-threaded, auto-retry)

hf upload-large-folder luckyrobots/ContentVault <local-folder>

Delete files remotely

hf upload luckyrobots/ContentVault . . --include="" --delete="<folder-name>/*" --commit-message "Remove <folder-name>"

---

Tips

• Resumable uploads: upload_large_folder and hf upload-large-folder are resumable. If interrupted, run the same command again.
• Don't use raw git push for large files — use the CLI/API instead. They handle LFS/Xet automatically.
• Write token required: Tokens default to read-only. Enable write access at https://huggingface.co/settings/tokens.
• 50 GB max per file on HuggingFace. Split larger files before uploading.
• Ignore patterns: Use --include and --exclude to filter what gets uploaded.