ldw: fix blinker cooldown using wrong time step

[Lidang-Jiang]

Summary

Fix recent_blinker timing bug in LaneDepartureWarning.update(). The blinker cooldown multiplied frame deltas by DT_CTRL (0.01s, 100 Hz control loop rate), but sm.frame advances at DT_MDL (0.05s, 20 Hz model rate). This made the intended 5-second cooldown effectively 25 seconds, causing LDW to remain suppressed far longer than intended after a lane change.

• Replace DT_CTRL with DT_MDL in import and usage (2 line changes)
• Add 7 unit tests validating correct timing at frame boundaries

Fixes #36126

Before (3 tests fail — cooldown incorrectly uses DT_CTRL)

============================= test session starts ==============================
platform linux -- Python 3.12.12, pytest-9.0.2, pluggy-1.6.0

selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_dt_mdl_is_used_not_dt_ctrl FAILED [ 14%]
selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_blinker_still_active_within_cooldown PASSED [ 28%]
selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_blinker_cooldown_expires_at_boundary FAILED [ 42%]
selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_blinker_cooldown_math_dt_ctrl_vs_dt_mdl PASSED [ 57%]
selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_right_blinker_cooldown FAILED [ 71%]
selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_no_ldw_when_speed_too_low PASSED [ 85%]
selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_no_ldw_when_lat_active PASSED [100%]

=================================== FAILURES ===================================
___________ TestRecentBlinkerTiming.test_dt_mdl_is_used_not_dt_ctrl ____________

    def test_dt_mdl_is_used_not_dt_ctrl(self):
      """Core regression test: DT_MDL (0.05) must be the multiplier, not DT_CTRL (0.01).

      With DT_MDL=0.05, 5s cooldown = 100 frames.
      With DT_CTRL=0.01, 5s cooldown would be 500 frames.

      At frame 101 after blinker:
        DT_MDL:  101 * 0.05 = 5.05s >= 5.0 => cooldown expired => LDW triggers
        DT_CTRL: 101 * 0.01 = 1.01s <  5.0 => cooldown active  => LDW suppressed (BUG)
      """
      ...
      ldw.update(101, model, cs_no_blinker, cc)
>     assert ldw.left is True, (
        "LDW should trigger after 101 model frames (~5.05s with DT_MDL); "
        "if this fails, DT_CTRL is likely still in use"
      )
E     AssertionError: LDW should trigger after 101 model frames (~5.05s with DT_MDL);
      if this fails, DT_CTRL is likely still in use

=================== 3 failed, 4 passed in 0.09s ====================

After (all 7 tests pass)

============================= test session starts ==============================
platform linux -- Python 3.12.12, pytest-9.0.2, pluggy-1.6.0

selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_dt_mdl_is_used_not_dt_ctrl PASSED [ 14%]
selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_blinker_still_active_within_cooldown PASSED [ 28%]
selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_blinker_cooldown_expires_at_boundary PASSED [ 42%]
selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_blinker_cooldown_math_dt_ctrl_vs_dt_mdl PASSED [ 57%]
selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_right_blinker_cooldown PASSED [ 71%]
selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_no_ldw_when_speed_too_low PASSED [ 85%]
selfdrive/controls/tests/test_ldw.py::TestRecentBlinkerTiming::test_no_ldw_when_lat_active PASSED [100%]

======================== 7 passed in 0.04s =========================

Test plan

• 7 unit tests with mock stubs (no full openpilot build required)
• Core regression test: verifies LDW triggers at 101 model frames (5.05s) — would fail with DT_CTRL
• Boundary test: verifies cooldown expires at exactly frame 100 (5.0s)
• Numerical demonstration: DT_CTRL produces 1.01s vs DT_MDL produces 5.05s at 101 frames

[github-actions]

Process replay diff report

Replays driving segments through this PR and compares the behavior to master.
Please review any changes carefully to ensure they are expected.

✅ 0 changed, 66 passed, 0 errors