Fix: lab_sim Tutorial 1 walk-through bugs

[rlpratt12]

Summary

lab_sim Objective cleanups surfaced by a Tutorial 1 walk-through on 9.2.1.

Problem

1. /masks_visualization doesn't auto-show. The ML Find Objects on Table Objective publishes masks but never calls SwitchUIPrimaryView, so a first-time user runs the Objective and sees nothing happen unless they know to manually switch a pane to /masks_visualization. The other ML Objectives in lab_sim (ml_segment_image, ml_find_bottles_on_table_from_image_exemplar) already auto-switch — this one is an outlier.
2. Deeply nested redundant Sequence wrappers in pick_all_bottles_with_apriltags.xml. The Release and Retract subtree had a Sequence containing a Sequence containing a Sequence containing a Sequence with the actual actions. Each intermediate wrapper had exactly one child, so the nesting carries no behavior — but it is genuinely confusing when a tutorial reader opens the tree.
3. Warm-up velocity_force_controller block at the top of Pick All Bottles with AprilTags. The objective starts with SwitchController → velocity_force_controller, zero_fts, SwitchController → joint_trajectory_admittance_controller before doing any picking, ostensibly to dodge a "cold start" on the first placement. The two controller swaps add a few seconds of dead time at the start of every run and the cost being guarded against is essentially zero — the same swap happens inside the place loop anyway.
4. Another redundant Sequence wrapper around Clear Snapshot / Look at Table / Take Wrist Camera Snapshot. Same shape as the Release and Retract problem above, just in a different part of the same tree.

Approach

Two commits:

1. fix(lab_sim): Tutorial 1 walk-through fixes
   • ml_find_objects_on_table.xml: append SwitchUIPrimaryView primary_view_name="/masks_visualization" to the top-level Sequence, mirroring the pattern used by the other ML Objectives in lab_sim.
   • pick_all_bottles_with_apriltags.xml: remove the three redundant Sequence wrappers inside Release and Retract. No reordering, no Behavior changes, just unwrap.
2. fix(lab_sim): Remove warm-up block and redundant Sequence wrapper
   • pick_all_bottles_with_apriltags.xml: delete the entire warm-up Sequence (SwitchController → velocity_force_controller, CallTriggerService /velocity_force_controller/zero_fts, SwitchController → joint_trajectory_admittance_controller).
   • pick_all_bottles_with_apriltags.xml: flatten the unnamed Sequence wrapping Clear Snapshot / Look at Table / Take Wrist Camera Snapshot into its parent.

Alternatives considered

• Adding a docs note saying "manually switch your view pane" instead of fixing the Objective — pushes the friction onto every reader and is inconsistent with the other ML Objectives.
• Keeping the warm-up block but moving it inside the loop's first iteration only — overcomplicates the tree for a problem that doesn't appear to exist in practice.
• Auto-deduping unary-Sequence nesting via a pre-commit hook — would catch other Objectives that have similar redundant wrapping, but is out of scope for this fix.

Validation

• pre-commit run --files src/lab_sim/objectives/ml_find_objects_on_table.xml src/lab_sim/objectives/pick_all_bottles_with_apriltags.xml — passes (prettier, Validate Objective favorites, codespell, mixed line endings, EOF).
• XML well-formedness verified by prettier and python3 -c "import xml.etree.ElementTree as ET; ET.parse(...)" on both files.

Manual verification

• Run ML Find Objects on Table in lab_sim and confirm the view pane swaps to /masks_visualization automatically once PublishMask2D returns.
• Run Pick All Bottles with AprilTags and confirm: the picking loop still places bottles in the tray; no regression versus the previous behavior on the first place (the warm-up was supposedly there to make the first place not be a cold start — verify the first place still succeeds without it).

🤖 Generated with Claude Code