commaai
Honda: Bosch gasonly pid
Honda Bosch has a built in PID tuner for brake, but does not have one for gas.
Even after adjusting gas for detectable conditions (incline, speed-based drag), these calculations are always bit off for undetectable conditions like tailwinds, weather/air density, etc. Thus a gas PID tuner is helpful for Honda Bosch.
However, the current openpilot PID tuner is only on overall accel. Any PID adjustment on gas then applies to brake, doubling up with the builtin Honda brake PID, and causing undesired braking strength until openpilot's PID adjustment winds down. Once it winds down, then you get undesired gas strength until it winds up again.
This pr allows for the PID tuner to only impact gas on Honda Bosch models.
The logic excludes Bosch B (radarless) since openpilot does not control the Bosch B gas pedal.
This is a replacement of opendbc pr2179 and opendbc pr35375, moved to a carcontroller PID at Shane's request. This is intended to be identical behavior to the prior prs.
Thanks to sunnyhaibin for python coding guidance.
New test routes:
Bosch C (2023 Honda Pilot): 9662a1dab70c9dc3/000000a6--a546f757e5
more coming from other cars in discord
Prior test routes from opendbc pr2179, that should have identical behavior:
Bosch C (2023 Honda Accord Hybrid): 15bd03c45fb5f401/0000000a--1805a6553f (test included https://github.com/commaai/opendbc/pull/2191)
Bosch C (2023 Honda Pilot): 9662a1dab70c9dc3/00000032--d0dac7d95b
Bosch C (2025 Acura MDX): ad9840558640c31d/00000013--893d36d514
Discord feedback:
I drove for five hours, and it was an incredible experience."
"worked well on the way to work today"
"Worked way better than the 0.9.8 branch I had last been using. Actually useable this time around. "
"once we can fix the rapid acceleration issue I think I’d daily op long" (retest in process after accel fix in https://github.com/commaai/opendbc/pull/2185)
"Acceleration and braking were exactly as they should be. I drove over to a construction zone to see how it would handle sudden starts/stops, and it was perfect." (test included https://github.com/commaai/opendbc/pull/2191)
Per our discussion, I'd like to try PIF control with drivetrain torque as the feedforward rather than acceleration.
Keeping this open but marking as draft, since it's now Plan B.
