MIGHTY is the default local planner; DROAN extracted to the asm_droan module (0.21.0-dev.11)

[andrewjong]

Summary

MIGHTY becomes AirStack's default local planner; DROAN leaves trunk as the asm_droan module.

• Default stack swap. full_default, lite_default and lite_offload_global:onboard now include the MIGHTY module launch (mighty_module.launch.xml: planner + acl-mapping voxel world model + NavigateTask bridge, CPU-only, lidar-fed) in place of droan_gl, and pin asm_mighty v0.1.4 in their modules.repos. Motivation: the judged obstacle-route comparison in the asm_mighty README — DROAN 0/5 and 0/10 (absorbing hovers, near-contacts) vs MIGHTY 5/5 with 1.59–1.65 m clearances behind the same controller.
• DROAN extracted. The six packages (droan_gl, droan_local_planner, disparity_expansion, disparity_graph, disparity_graph_cost_map, cost_map_interface) move to castacks/asm_droan (git filter-repo, 65 commits of history preserved; tagged v0.1.0; registered as the droan module). The droan_gl GL link deps leave Dockerfile.robot and become the module's tier-1 apt deps; the trunk image gains MIGHTY's nlohmann-json3-dev header dep so the default stack runs on the base image with no composed module layer (asm_mighty v0.1.4 declares no deps).
• Stacks. New full_droan = the previous default topology (GPU DROAN) pinning asm_droan — it inherits the old full_default wiring baseline (identical graph; only the H1 stack name changed). full_droan_cpu and full_macvo pin asm_droan. full_mighty is folded into full_default.
• airstack up reconciles stack pins. A stack's modules.repos is honoured at launch: missing pins are added + synced, differing local pins are kept and named, uncomposed dependency layers are built (module lock --build, unless AIRSTACK_NO_IMAGE_BUILD=1). --dry-run/--config-only only report; AIRSTACK_NO_STACK_MODULE_SYNC=1 opts out. Without this the module-pinned default stack would fail on first run.
• Docs/skills/tests. Local-layer pages rewritten (planner comparison table, world-model pairing), stacks/autonomy-modes tables, walkthrough, module/stack docs, skills, mkdocs nav (5 package READMEs out, full_droan + droan in), registry fixture + regenerated catalog, release notes, VERSION → 0.21.0-dev.11. New contract tests pin the dry-run reconcile report.

Companion PRs (merge the registry one first — the docs deploy regenerates the catalog from the live registry): castacks/airstack-modules-index (droan + full_droan entries, full_mighty removed, mighty re-pinned to v0.1.4).

Validation

Check	Result
pytest tests -m unit (stack layout, single-locus lint, docs catalog --check, launch-intent dry-run, CLI help, new reconcile tests)	404 passed, 11 skipped
tools/validate_module.py asm_droan / asm_mighty; registry validate_entry.py	valid (9 registry entries)
mkdocs build	0 warnings
In-container colcon build (robot 0.21.0-dev.9 base + the new apt deps, trunk ws with the six packages removed + asm_droan/asm_mighty overlaid; --packages-up-to the module packages)	21 packages finished, 0 failed (51 packages discovered, none duplicated; 12 min 47 s)
airstack up --dry-run (default / full_droan_cpu)	reports the pins, touches nothing

Version note

The module compat floors (>=0.21.0-dev.10) and the module READMEs were written before #423 landed on dev.10; the swap itself ships in dev.11. dev.10 is a catalog-only sync nobody targets, so the floors are left one dev release permissive rather than re-tagging asm_droan v0.1.0 / asm_mighty v0.1.4.

Not done in this PR (needs a sim session)

• full_default (MIGHTY) and lite_default have no committed wiring.md — the stale DROAN baselines were dropped; the snapshot test passes with the bootstrap instruction until airstack test -m wiring --stack full_default --sim isaacsim --num-robots 1 is run and its observed file committed. (full_mighty shipped the same way.)
• The 13 droan/* keepalive rows stay in topic_keepalive_node.py: removing them drifts the full_droan / full_droan_cpu / full_macvo goldens, which must be regenerated in the same change (tracked in asm_droan TRUNK_REMOVAL.md §4).
• No end-to-end flight of the new default was flown here (GPU shared with running experiments); the MIGHTY-on-0.20.x closed-loop flights and the judged campaign are the module's evidence.

🤖 Generated with Claude Code

[github-actions]

Test Metrics — d86c4752380996693e72f8b77b52a3ecaf112420

This was a unit/build-only run. Simulation regression comparison does not apply.

system.test_build_packages

Pass rates

Test	Pass	Fail	Skip	Rate
test_colcon_build_gcs	0	1	0	0%
test_colcon_build_ms_airsim	0	1	0	0%
test_colcon_build_robot	0	1	0	0%
test_colcon_test_robot	0	1	0	0%

Metrics

Test	Metric	Value
test_colcon_build_robot	duration_s	0.82s
test_colcon_test_robot	duration_s	0.63s
test_colcon_build_gcs	duration_s	0.631s
test_colcon_build_ms_airsim	duration_s	0.619s