[trung791997]

Stacked on #3 (the Bosch-A parser port). Review only the top commit, 427f198.

This brings Testing's Bosch-A actuator code from Biscuit-Labs/openpilot Honda-PTM-Testing into nrdr. On qualifying Bosch-A cars it replaces nrdr's gas and wind factors and its Bosch extra-brake controller. It is always on, with no toggles. Other cars are unchanged.

Ported from Testing: bosch_longitudinal.py (including BoschABrakeEdge), brake_onset.py, bosch_low_speed_brake.py and bosch_gas_learner.py.

• brake_onset, bosch_low_speed_brake and bosch_gas_learner are byte-identical to Testing.
• bosch_longitudinal only adds star7_params and update_mvl_gas. No number or logic from Testing changed.

When it runs

• The car is a Bosch-A platform with the Bosch-A radar active (radarUnavailable False) and openpilot longitudinal. A Civic with its radar turned off keeps nrdr's path.
• Cars using the alternate Bosch radar (BOSCH_ALT_RADAR) are not included, because nrdr's Bosch-A list leaves them out. Testing's list includes them. The 2019 Civic is not affected.

Gas learner

• Testing's CivicGasLearner runs on HONDA_CIVIC_BOSCH only, as on Testing. Other Bosch-A cars get Testing's fixed gas and wind factors.
• On the Civic it also replaces MVL's learner when MVL is on (mvl_longitudinal.py, update_mvl_gas). The car owner reported that MVL's learner over-revs on hills.
• With MVL on, the Civic learner works on MVL's gas table, not on Testing's 750 table.

Stopping

• On this path the stop counter counts the stopping state alone, as Testing does (Testing carcontroller l.257).
• nrdr's extra accel <= 0 condition stays for every other Bosch car. That condition was there because nrdr's stopping ramp can still send positive acceleration. On this path the ramp is not used.
• Result: if the planner says "stopping" but asks for positive acceleration, standstill stays held, where nrdr would release it.
• test_bosch_stopping.py has updated Civic expectations that follow from this rule. The sent acceleration in the last case was recorded from a run.

hondacan.create_acc_commands gains an optional braking=None argument. Callers that leave it out get the old behaviour.

Tests

• Static only. Nothing here is replay- or road-validated on nrdr.
• Added test_bosch_longitudinal, test_brake_onset, test_bosch_low_speed_brake and test_bosch_gas_learner from Testing.
• Added test_bosch_a_long_wiring, 11 tests covering the gate, the replaced controllers and the MVL learner swap.
• opendbc/car/honda/tests: 1267 passed, 43 failed, 3 skipped. The 43 failures also fail without this change: 42 in test_mvl_priority.py and 1 in test_mvl_sync.py.
• ruff is clean on the changed files.

🤖 Generated with Claude Code

[nrdr]

Reconciled the two repositories: opendbc's nrdr-development now points directly to 427f1986ed241f111d106cc72cdefbda5efe5f7f, the exact revision already selected by merged nrdr/openpilot#27. The fast-forward includes both this actuator commit and its parser parent (#3), with no merge conflicts or source edits.

PR #2's Civic Bosch/Insight 3840 torque limits remain present. The older radar PR #1 is already ported and superseded. This reconciliation does not add replay/road-validation evidence or deploy anything to a vehicle.