[TomCC7]

Superseded by #4960, which is based directly on main and excludes the learning stack.

Stacked on #4900. Target this integration branch while the learning stack is open; retarget to main after it lands.

YAM, OpenArm, and MakerMods Metal currently let the coordinator own their Damiao adapters directly. Add standard connection modules and dedicated *-connection-control blueprints so the connection owns CAN, activation, state publication, deadman holds, and shutdown. Follow the connection-module approach in #4958 while retaining the existing adapters, MIT gains, gravity compensation, and normalized gripper openings. Existing blueprints and learning compositions are unchanged.

The control models expose logical 0–1 gripper coordinates without changing physical finger geometry. Reuse the model-backed gripper support from #4939 (27d3753ca1) so coordinator gripper tasks work over connection streams.

Validation:

• 56 new mocked tests cover command mapping, partial holds, gains, limits, grippers, coordinator routing, estop, deadman, feedback faults, CAN selection, and cleanup.
• 208 applicable regression tests and 5 blueprint-generation tests pass. Four marked tests were excluded; one unchanged Metal adapter test is blocked by the local can-motor-control 0.0.8 environment, which lacks the required >=0.0.11 MotorProfile API.
• Mypy passes for the 11 changed production files; pre-commit passes. Cached official models resolve to 7 Metal, 7 YAM, and 16 OpenArm control coordinates.
• No physical robot actuation or hardware acceptance testing. No dependencies were installed or upgraded.

Run commands and interface details: Damiao connections.

[github-actions]

go2 replay realtime (arm64)

Details

Benchmark suite	Current: 735b9fb	Previous: d2fb73c	Ratio
peak memory	1740.656 MB	1737.02 MB	1.00
peak threads	376 threads	371 threads	1.01
network (transport)	2725.82 MB	2733.743 MB	1.00
disk write	14.285 MB	1.699 MB	8.41
instructions	138.097 G	136.49 G	1.01

This comment was automatically generated by workflow using github-action-benchmark.