Space Lab 2

[west siren]

https://github.com/commaai/openpilot/tree/sl2

<@&1386399863293345912>

[lusty ruin]

Let’s get it @west siren

[plain hornet]

custom branch this time?

[north kayak]

Give it up for another model iteration wooo

[west siren]

we're waiting for @wild wasp to add it for us

[plain hornet]

ah

[west siren]

https://tenor.com/view/well-waiting-impatient-in-a-hurry-madison-gif-8657265

[wild wasp]

Nah it’s a pr branch still

[wild wasp]

Just triggered workflow

At target

[west siren]

get me some smackin seeds lol

[lusty ruin]

Ill test this one out when it drops

[west siren]

ill wait for you to give me a go to test

[lusty ruin]

I ain’t scared

[west siren]

i mean someone got take one for the team

[lusty ruin]

I’m gonna try all the curves

[wild wasp]

https://github.com/sunnypilot/sunnypilot/actions/runs/16534755642/job/46767127558

Monitor here to know when to refresh

Looks like it’s finishing up though

It’s done @lusty ruin @west siren

[lusty ruin]

I need a webhook

[west siren]

I’ll try it out in lil bit

[lusty ruin]

Oh you added a refresh model list button

[wild wasp]

@rancid tinsel ☺️

[lusty ruin]

Something different about steering

Feels loose at the beginning of the turn

[wild wasp]

Rip

[lusty ruin]

It gets tight though

Its weird

[west siren]

Usually it’s the other way around 😂

[lusty ruin]

And I got a torque warning on a lane change

[crimson ivy]

Lat is pretty good so far

[junior cloud]

Space Lab 2 been doing her kegel exercises.

[lusty ruin]

It wasn’t bad for LAT but I think I like v1 more

And by think I mean I know

[plain hornet]

oof

v1 better then ig

[maiden kraken]

so you recommend don't bother to test this one?

[lusty ruin]

I say test it because I need a sanity check on the turns

[tranquil kraken]

Lat seems more centered so far

[maiden kraken]

what year and model of car?

[tranquil kraken]

'17 rav4

[crimson ivy]

Lat good, it does seem to stop for red lights and it went go on 1 of 2 green lights. Long idk, it was at night and all the models seem to be giving me rubberbanding. Not much traffic to evaluate long

[crimson ivy]

I just realized I did the whole 100 mile test with NNLC off. So very good lateral

[junior cloud]

Was that with very sharp turns or mostly interstate?

[mild temple]

this ones decent

[lusty ruin]

I’m gonna test long today

[crimson ivy]

highways, there were some hard curves but none of it was low speed

[junior cloud]

Personally I don’t really notice a lot of difference with highway speeds with NNLC on unless I’m on some super curvy scenic route.

Where NNLC really shines for me is low speed turns and curves.

[crimson ivy]

low speed was good too, but i don't have so many curves nearby

[crimson ivy]

it also likes to follow leads into right turn lanes

[junior marlin]

Recentering is slower than SP and DTR5 but better than any TR in my experience in city driving

[quiet kraken]

I like the lane centering in SL2 better than the rest of the models that were released this month. This one stops smoothly when there is a lead already stopped, but comes up too fast to the intersection when you don't have a lead car there.

Once today I used the brakes to get a smooth stop for a red light, then while holding the brake pedal re-engaged it. It sat for about two seconds and then began to move like it was going to go through the red light. I had to stop it with the brake again.

[lusty ruin]

did you try the newer SL2?

[quiet kraken]

? newer SL2? I was using the newest one I saw; this was about 7 hours ago now.

Here's the clip from the Go-on-Red event. #sunnypilot-clips message

Oh. Now I see a post about SL2 v2. I'll give it a test either later tonight or tomorrow. I reset calibration when I change models and drive until it says the Live Delay is calibrated and live Torque is calibrated. Today I think I observed that the live Delay doesn't begin unless SP is controlling both lat and long. It didn't start calibrating the live delay while only MADS was in use. A 25 minute drive can get them both calibrated.

[junior marlin]

Live delay also keeps adjusting as you continue to drive after it’s calibrated. I have seen .02-.04 swings over time.

[quiet kraken]

I noticed that too. I keep a journal and write a short note in it each time I stop a route. I've started recording the Live Delay and saw it change after it was calibrated just as you mention.

[wild wasp]

Resetting calibration isn’t needed unless you’re changing different vehicles

[quiet kraken]

I know you've said that before, and other people have too, but I was thinking that the Live Torque and Live Delay might benefit from a fresh start. If they both continuously re-calibrate then eventually the new model will get the values that it seems to need. I've seen that different models get torque values and delay values that are not close to what other models settled into. By resetting calibration I can see when the new model has its own numbers.

If a model is using the delay that was from a different model, it will behave differently on a curve, not following it exactly and maybe entering it late or early depending on which way the delay is from where it would like it to be.

[wild wasp]

Well the delay is from your vehicle/software. it’s estimated vehicle steering delay. Not 100% accurate though as we see it changes widely depending on road, friction, heat etc

[quiet kraken]

But different models end up with different delays (same car, mine).

[wild wasp]

Sure and I agree. But I was just noting that it isn’t the model that live delay comes from. It’s a separate process that runs

[quiet kraken]

Well, wherever it comes from, I see different delay values after the models calibrate. TR16 got 0.315, SLMv1 got 0.294 and 0.313, SL2 got 0.434 and 0.427, SL2v2 got 0.300. When the toggle was first put in to let users choose a delay I tried values close to the 'default' and saw that small changes had noticeable effects. So now, seeing these calibrated numbers that run from 0.294 to 0.434 I thought it was because of the model. Are you saying that's not the case, these values are just my car's mechanical parts?

[dense minnow]

They change based on a lot of factors. If you drive with openpilot enable practically all the time, it won’t really narrow down to anything meaningful. But if you drive the same road at the same speed in the same weather at the same time of day, you’ll probably get the same value to populate. In the end, the value is gonna change, but it’ll be close.