GM Bolt CC-only: drive stock set speed from planned speed, not PID accel

[OVR92]

GM Bolt CC-only: drive the stock set speed from planned speed instead of PID accel

Problem

On CC-only Bolts (no ACC, no pedal interceptor) the cruise-button longitudinal constantly taps the set speed up and down, both with no lead and behind a steady lead.

Two root causes:

1. The CC-only longitudinal tune (kpV = [0, 5, 2] with a 0.9 m/s deadzone) was written for the retired speed-error PID. LongControl.update() now runs a PID on acceleration error and never applies the deadzone (it is only referenced in the legacy update_old_long path). On the accel loop, kp = 5 turns ±0.1 m/s² of aEgo ripple into ±0.5 m/s² of command.
2. create_gm_cc_spam_command projected a set speed from that accel command (v_ego * 1.01 + 3 * accel) with no cap at the cruise set speed and no notion of target speed. The stock cruise is itself a slow speed servo, so a noisy accel signal quantised into 1 mph steps at up to 5 taps/s limit-cycles.

Change (Bolt CC fingerprints only)

• gmcan: _bolt_cc_setpoint_button targets the planner's speed 2.5 s ahead (actuators.speed, already published by controlsd), low-pass filtered (3 s up / 1.5 s down) and capped at the cruise set speed. Taps need a 0.8 mph difference and wait 1.5 s in steady state. A planned-speed drop > 1 m/s or a decel request ≤ -0.5 m/s² bypasses the filter and taps down every 0.25 s. No taps below the stock 25 mph minimum. Removes the previous reverse-confirm helper.
• interface: Bolt CC kp 5 → 0.5 so the accel command tracks the planner. Other CC-only cars keep [0, 5, 2].
• carcontroller: passes the HUD set speed into the spam logic and clears the target filter when spam is inactive.
• starpilot_following: +0.4 s follow time on cruise-button Bolts, which can only slow by coasting once the set speed is lowered.

Tests

• New: set speed holds with no lead; cruise speed caps the set speed; small target changes are filtered; sudden planned-speed drops bypass the filter and tap fast; rising targets are slower than falling; no taps below the stock minimum; cancel when far below engage speed; spam command respects the tap interval.
• Boundary: test_other_cc_only_cars_keep_existing_long_tune (Equinox CC keeps [0, 5, 2]) and the existing test_non_bolt_cc_redneck_keeps_existing_setpoint_selector (Equinox CC keeps the accel-projection selector).
• Updated test_non_volt_cc_redneck_spam_stays_on_powertrain_bus fixture for the stateful Bolt path.
• Run: the GM test modules (test_gm.py, test_gmcan.py, test_carcontroller.py, 255 tests) pass on a Debian 12 aarch64 host. Ruff 0.13.1 on the changed files reports only two pre-existing E501 lines that this PR does not touch.

On-road testing

2023 Chevrolet Bolt EUV without ACC or pedal interceptor, comma 4, mostly highway. Set speed holds still with no lead, taps every several seconds behind a steady lead, and drops quickly when a lead brakes. Reported as a large improvement over the previous behavior.

Note for reviewers

The dead-deadzone finding applies to every CC-only GM car in that tune block (Equinox CC, Trailblazer CC, XT4 CC, Volt CC). This PR deliberately does not change them; they still run the accel-projection selector, and I only tested the Bolt.

🤖 Generated with Claude Code