opendbc icon indicating copy to clipboard operation
opendbc copied to clipboard

GM: Silverado/Sierra better longitudinal tune

Open morrislee opened this issue 6 months ago • 14 comments

This PR proposes reducing stopAccel from -2.0 to -0.39. The current value introduces overly aggressive braking, as shown in the comparison log where aEgo sharply spikes to -2.0 m/s², diverging from planner intent.

In contrast, a value of -0.39 closely matches stock ACC behavior and aligns well with planner targets, producing smoother and more natural stops.

📊 Three comparative logs are included:

Stock OpenPilot (-2.0) — aggressive decel with overshoot dc7716b32bf25574/00000029--3581c1f42c/2 (OP Long stopAccel = -2.0) image

OEM ACC — smooth taper, no jerk dc7716b32bf25574/0000002a--a6b57eae9a/1 (Stock Long) image

Tuned OpenPilot (-0.39) — matched decel to planner and actuator, no overshoot dc7716b32bf25574/0000002c--ee2c92ef33/1 (OP Long stopAccel = -0.39) image

🧪 Delay Compensation Evaluation To validate the effective longitudinal actuator delay, we compare the planned acceleration across various delay indices (6–9) with the actual actuator command.

📈 Full-segment analysis (shown below) demonstrates: dc7716b32bf25574/0000002c--ee2c92ef33/1:6 image

This confirms that the system behaves as if it has ~0.625s delay. Combined with the stop behavior analysis, this supports reducing stopAccel to -0.39 to avoid overshoot due to delay-compensated controller output.

  • Dongle ID: dc7716b32bf25574

I hope @sshane can sanction this slightly improvement :D

morrislee avatar Jun 03 '25 02:06 morrislee

We should match the stock system, did you check what it does coming to a stop? Can you get a route? Also need to check this works on steep hills

sshane avatar Jun 03 '25 02:06 sshane

OEM ACC — smooth taper, no jerk dc7716b32bf25574/0000002a--a6b57eae9a/1 (Stock Long) image stock acc is provided in the second graph

morrislee avatar Jun 03 '25 03:06 morrislee

Again, we need to see what stock ACC commands, not the effect. That plot tells you nothing.

sshane avatar Jun 03 '25 03:06 sshane

Again, we need to see what stock ACC commands, not the effect. That plot tells you nothing. does this help? image

morrislee avatar Jun 03 '25 03:06 morrislee

Stock ACC image

OP Long with -0.39 image

morrislee avatar Jun 03 '25 03:06 morrislee

Stock image

-0.39 image

morrislee avatar Jun 03 '25 03:06 morrislee

Stock image

-0.39 image

morrislee avatar Jun 03 '25 04:06 morrislee

@sshane does the graphs help? I am not sure what I am looking for, could use a hand here

morrislee avatar Jun 03 '25 04:06 morrislee

@sshane

We should match the stock system, did you check what it does coming to a stop? Can you get a route? Also need to check this works on steep hills

Not many hills here, this is the steepest ones I can find to do stopping behind lead cars

Uphill dc7716b32bf25574/0000002e--781a6d659e/29:31 image

Downhill dc7716b32bf25574/0000002e--781a6d659e/45 image

Downhill dc7716b32bf25574/0000002e--781a6d659e/51 image

morrislee avatar Jun 04 '25 02:06 morrislee

Can you get one other user with this car to try these changes and post a route?

sshane avatar Jun 06 '25 04:06 sshane

Is this safe for the cars that fingerprint as Silverados that aren’t Silverado?

garrettpall avatar Jun 06 '25 17:06 garrettpall

Is this safe for the cars that fingerprint as Silverados that aren’t Silverado?

if it works for a large truck, I dont expect it would be worse for smaller cars. FYI, this never contributed to a shorter stopping distance that I can tell

morrislee avatar Jun 06 '25 19:06 morrislee

This PR has had no activity for 60 days. It will be automatically closed in 7 days if there is no activity.

github-actions[bot] avatar Sep 21 '25 02:09 github-actions[bot]

This PR has been automatically closed due to inactivity. Feel free to re-open once activity resumes.

github-actions[bot] avatar Sep 28 '25 02:09 github-actions[bot]