[Nick Hehr (HipsterBrown)]

Summary

This PR adds a new sequential workshop, Teach an xArm6 to Stack Blocks, for an upcoming hackathon. It walks through end-to-end robot learning. A person teleoperates an xArm6 follower with an SO-ARM101 leader, and Viam records each demonstration. Participants train an ACT (Action Chunking with Transformers) policy with LeRobot on a cloud GPU, then deploy it on the machine. The leader and follower are different arms on purpose, to show mixed hardware on one platform.

flowchart LR
    T["Teleop<br/>SO-ARM101 → xArm6<br/>(Cartesian teleop)"] --> R["Record<br/>capture control sensor<br/>+ capture-control webapp"]
    R --> C["Curate<br/>review:keep →<br/>sequence dataset"]
    C --> Tr["Train<br/>custom ACT script<br/>(cloud GPU)"]
    Tr --> D["Deploy<br/>viam-labs:vla<br/>policy + controller"]

Loading

 docs/tutorials/
 ├── pick-and-place/
 ├── so-arm101-palletizing/
+└── xarm-act-stacking/
+    ├── _index.md                        # overview, hardware, plan-your-day table
+    ├── _phase-template.md               # headless
+    ├── 01-platform-mental-model.md
+    ├── 02-configure-the-cell.md         # fragment, teleop mapping, home poses
+    ├── 03-record-demonstrations.md      # capture control + webapp
+    ├── 04-review-and-build-a-dataset.md # keep/discard, sequence dataset
+    ├── 05-train-a-policy.md             # milestone one
+    ├── 06-deploy-the-policy.md          # milestone two
+    ├── 07-evaluate-and-iterate.md       # optional
+    └── 08-wrap-up.md
 .htmltest.yml                           # ignore pending companion repo

The pages use the existing workshop format: front matter registers the workshop, and the pages use the checkpoint and workshop-nav shortcodes. This PR changes no layout or shortcode files. Every page is draft: true until the dependencies below land. The capability tags are all existing ones, so docs-health needs no change.

Evidence

• Before: /tutorials/ shows two workshop cards (pick-and-place, so-arm101-palletizing).
  After: a draft build (hugo --buildDrafts) shows a third card. The nav chain renders overview → Phases 1–7 → wrap-up, each page shows "Phase N of 7", and no phase pages leak into /tutorials/all/. A production build excludes the draft pages.
• prettier 3.2.5, markdownlint, vale, scripts/frontmatter_rules.py, and make build-prod all pass. Local vale was 3.26.0, not the 3.12.0 that CI pins, so CI's run is the source of truth.
• Each page passed a spec-compliance and quality review against its sources: the vla-inference-service, smolvla-so101-training, and capture-control-webapp READMEs and code, and the teleop-cartesian doc. A final pass reviewed the whole branch for cross-page consistency and safety.

Before publishing (TODO(dep) markers)

Blocking milestone two:

• viam:ufactory:gripper needs a {"set": N, "wait": false} option on DoCommand. Today set polls until the jaw settles, up to 10 seconds, which stalls the 10 Hz control loop (06:266). Add optional non-blocking gripper "set" DoCommand viam-modules/viam-ufactory-xarm#145 adds it. Once it's released, add "write_args": {"wait": false} to the Phase 6 controller's gripper config and remove the "arm freezes at a gripper change" troubleshooting row.
• An ACT checkpoint loads and runs through viam-labs:vla:policy. Verified on vla 0.2.1 on an Intel NUC: the policy reached ready, its specs were correct, and one infer returned 100 actions in 0.22 s on the CPU.

Teleop module extraction:

• Module ID, model name, and attributes (02:193), and a docs link (02:239).
• What rejected counts (02:269).
• Recheck the minimum viam-server version, currently v0.119.0 (_index:103).

ACT training script and companion repo (viam-devrel/xarm-act-stacking):

• Publish the training script. It's public as devrel:act-policy 0.1.0, a general ACT training script, with the Robot policy task type. Its arm, gripper, and cameras arguments are required.
• Make the companion repo public. It's private for now.
• The script name, the container, the argument names, and the defaults: steps 50,000 and image_size 0. These are measured on an A100, at 0.081 s per step. Phase 5 passes the xArm arguments explicitly.
• The script builds its input features from the dataset, so the feature keys follow the camera names.
• Mat zone count (07:27).
• Public fragment ID, if one is published (02:49).
• cell.json must set the teleop gripper_invert: true and leader_tool_offset_mm, because Phase 2 tells readers they're already set.

Version pins:

• viam-labs:vla 0.2.1, which matches the training container's LeRobot commit.
• viam:capture-control (03:27).

Verify on hardware:

• xArm gripper open and closed values, about 840 and 2 (05:54, 06:143).
• state_dim and action_dim of 7 on an xArm checkpoint (06:135). They were 6 and 6 on the 5-joint SO-101 checkpoint, as expected.
• xArm6 joint_limits_degs (06:192).
• Measured actions_per_chunk (06:214). On a NUC's CPU, the floor is 3.
• RealSense color-only attribute (02:118).
• Cameras with different resolutions. The script expects them to work, but the validation dataset's cameras were both 640x480.

Platform checks:

• The capture override keeps the gripper's additional_params.docommand_input (03:94).
• Saving a pose to a switch defined in a fragment works (02:256).
• The "All sequences" dialog can filter by tag (04:76).
• The hosted webapp URL is live (03:125).

Also:

• Replace the ASSET placeholders with photos, screenshots, and diagrams.
• Do a full dry run on real hardware to set the time estimates.
• Set draft: false.

Merge Danger

Door: two-way

All pages are drafts, so merging publishes nothing. Reverting is a plain revert.

Blast Radius: none

The only change outside the workshop directory is one IgnoreURLs entry in .htmltest.yml, for the companion repo that doesn't exist yet.

🤖 Generated with Claude Code

https://claude.ai/code/session_012FAJrDL3amKcRmphzeVnyb

[netlify]

✅ Deploy Preview for viam-docs ready!

Name	Link
🔨 Latest commit	4d680f4
🔍 Latest deploy log	https://app.netlify.com/projects/viam-docs/deploys/6aeba6490cd10ba44d23ffa5
😎 Deploy Preview	https://deploy-preview-5493--viam-docs.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.
Lighthouse	1 paths audited Performance: 31 (🔴 down 3 from production) Accessibility: 100 (no change from production) Best Practices: 100 (no change from production) SEO: 92 (no change from production) PWA: 60 (no change from production) View the detailed breakdown and full score reports

---

To edit notification comments on pull requests, go to your Netlify project configuration.