GMC Acadia Denali: cruise fault immediately after releasing the accelerator when engaged

[Verylukyguy]

Describe the bug

When OpenPilot is already engaged and you begin accelerating with the Disengage on Accelerator toggle turned off, you get a Cruise Fault: Take Control Immediately RSOD immediately upon releasing the accelerator.

This also happens when you engage OpenPilot while accelerating and then the accelerator is released.

I provided a route with several examples.

What hardware does this issue affect?

comma two

Provide a route where the issue occurs

e38a02cae1ef1e0c|2022-04-05--11-10-35--0

openpilot version

Comma Master 0.8.14 @ f0ac808

Additional info

This behavior is the same when the Disengage on Gas toggle is on, see route: e38a02cae1ef1e0c|2022-04-05--11-02-43--0

[sshane]

Can you try this branch out? https://github.com/commaai/openpilot/tree/gm-cruise-faults

I think it may be related to how we now use longActive instead of enabled for a couple of signals related to ACC (changed in PR #23859).

When you press the gas, we set ASCMGasRegenCmd->GasRegenCmdActive to False which causes AcceleratorPedal2->CruiseState to fall, then when you release the gas we start commanding gas/regen immediately while the CruiseState takes a frame to rise again, so maybe that's the cause of the faults? Let me know how the branch goes! @Verylukyguy

[Verylukyguy]

Can you try this branch out? https://github.com/commaai/openpilot/tree/gm-cruise-faults

I think it may be related to how we now use longActive instead of enabled for a couple of signals related to ACC (changed in PR #23859).

When you press the gas, we set ASCMGasRegenCmd->GasRegenCmdActive to False which causes AcceleratorPedal2->CruiseState to fall, then when you release the gas we start commanding gas/regen immediately while the CruiseState takes a frame to rise again, so maybe that's the cause of the faults? Let me know how the branch goes! @Verylukyguy

This made it worse; now it faults when the accelerator is pressed while engaged.

[sshane]

It looks like this might have always been an issue. We probably need a stock route of an ICE car to see what the ASCM sends to avoid faults as I tried duplicating what it does for the Volt, but that made the issue worse. Doesn't seem to be an issue on the Volts, which is what @twilsonco said the GM port started with.

[Verylukyguy]

It has been an ongoing issue on stock OpenPilot, but it is very strange that the Volt code didn't work, because I am usually on some version of Tim's fork and I do not get those errors. I will attempt to get you some factory GM routes to analyze. Thank you.

[Verylukyguy]

I think that I may have actually found a reference to this issue in #348

[twilsonco]

This happens in my '18 Volt with C3 as well; not just Acadia.

Importantly, this does not happen in my fork (tw-0.8.9 branch) GitHub.com/twilsonco/openpilot, or in the opgm fork ('dev' branch) gihub.com/opgm/openpilot.

steps to reproduce on apparently any GM vehicle:

1. in master, disable the disengage on gas toggle
2. Drive, engage
3. Press gas; lateral stays engaged, but you'll have long control with the gas pedal
4. release gas pedal

Expected behavior: OP resumes long control
Observed behavior: RSOD "Cruise Faulted Take control immediately". This resolves within seconds and you can reengage.

[twilsonco]

I'll attempt to fix and submit a PR

edit: nothing prints in tmux to accompany the fault.

[twilsonco]

I'm stumped for the time being. Reaching out to @JasonJShuler and @qadmus for help. Here's from discord:

steering stays engaged while you press the gas, but then ACC faults right after you release the pedal. In these plots I got from Shane of stock GM cruise doing it's no-disengage-on-gas thing, I notice that, at t=148s when gasPressed goes to 1, the acc keeps sending a non-"zero" GasRegenCmd for ~1s before it drops to "zero", then it comes back to the 1900 value ~0.5s before the gas is released.

I thought maybe this meant there had to be a non-"zero" gas command present when op-long resumes control, but I don't think that makes sense now...
More likely, we're missing a frame in the ACC commands in one of two ways. When the user releases the gas, op sends the first true GasRegenCmdActive signal either:

1. a frame too early, overlapping with the gas pedal and causing a fault, or
2. a frame too late, resulting in missing frames and causing a fault.