openpilot
openpilot copied to clipboard
commaai
HKG: Can-fd resume from standstill improvement
Description Resume from standstill fails to work 1/5 times in a given drive in my hda1 EV6 (and others as well). To solve this problem I have employed the same technique as used in non-canfd hkg cars. I propose sending 5 "resume" button presses at a time. I attempted sending 25 button presses as is currently done with non-canfd cars, but this resulted in a communications error. 5 button presses appears to improve the resume from standstill (no failures in 3 days of driving) and does not give the communications error.
Verification This fix has been employed for a few days now by a few users in a fork created specifically for hda1 ev6 and I5 cars (hoomoose:hda1ev6i5) since master does not have the prerequisite resume from standstill fix needed for hda1 ev6 and I5. In three days of driving with this fix, I have not had a failure of resume from standstill (yet...). However, this fix would affect all can-fd cars with the cruise_buttons signal and can/should be tested by hda1 and hda2 cars.
Prerequisite For testing on an HDA1 car, the hda1 fix is a prerequisite: https://github.com/commaai/openpilot/pull/26069
I had my first failure with this method, so it is not perfect. Just thought I would report this.
@hoomoose Have you tried something similar to this PR?
- https://github.com/commaai/openpilot/pull/25579
I just tested out the solution offered by Sunnyhaibin, and have committed the changes. In one day of driving through rush hour traffic, I had no errors and 100% resume rate. I would estimate success was 14/14. I do not think a conclusion was reached whether to merge the previous PR, but it certainly seems to cause no negative issues with an EV6 and would appear (with limited data from one can-fd car) to offer success. What would be a next step - have more cars try this out?
I just tested out the solution offered by Sunnyhaibin, and have committed the changes. In one day of driving through rush hour traffic, I had no errors and 100% resume rate. I would estimate success was 14/14. I do not think a conclusion was reached whether to merge the previous PR, but it certainly seems to cause no negative issues with an EV6 and would appear (with limited data from one can-fd car) to offer success. What would be a next step - have more cars try this out?
@hoomoose Provide the route(s) with the resume tests should be sufficient for review.
Now that HDA1 resume is in master, can you try master for a bit and see if you still have problems?
In driving through downtown to and from my rec center on the new Master with no modifications from me (which is awesome by the way!), I had the following results. To gym 4/6 successful resumes (Route 68d6a96e703c00c9|2022-10-19--16-28-36) From Gym 7/8 successful resumes (Route 68d6a96e703c00c9|2022-10-19--17-49-42) (and 1 false positive where the car resumed but the car in front wasn't moving, but that's a success in my book.) Overall successful resume rate 11/14. I am happy to keep driving as is to get more data, but I would say this has been typical of my experience all along before trying to do improvements. Others who post (can-fd with hda1 and hda2) seem to have similar experiences of the ~15 -25% failure rate. Do you have a similar experience with your EV6? I will also try this fix again after re-basing with current master to see if I have the same success rate as I previously posted and I will post routes.