-
Notifications
You must be signed in to change notification settings - Fork 9.1k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Tall vehicles: calibrate vlead from model error #26926
Conversation
I think it's best not to correct lead_a, can we just do v? |
if v_ego > 0 and vision_v_ego > 0: | ||
vision_velocity_factor = v_ego / vision_v_ego | ||
corrected_v_lead = lead_msg.v[0] * vision_velocity_factor |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Might make more sense to copy what longitudinal_planner does, which is simply get an error to subtract
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
should be a scale rather than a subtract, if there is an error, then it might apply to a static lead no?
suggested by Harold for consistency to scaling vlead
My 2018 Acadia is considered a Mid-Sized SUV and the camera height is 53" or 1.35 meters. This is only slightly higher that the standard 1.22 meters. I've heard that increasing the height might mitigate some of the lane hugging. Is this true? |
this calibration will not have any effect on lateral performance, at least, not supposed to. it is strictly scaling the speed leads are traveling and also assist on how fast e2e think the car is going. |
Interesting. Skip did a lot of trial-and-error testing when the Silverado was first ported. His camera height was 1.62 meters and it seemed to cause wandering and hugging. At least, that is what he thought. |
been driving with harold's PR #27493 seems to be very similar in performance, will close this PR as harold's approach is a bit cleaner and safer |
Description
When using vision only longitudinal control from OpenPilot (more prominent on taller vehicles), it estimates the world objects moving too slow overall including itself #26799. This is verified from speed reported from the model is off from wheel speed, we used this info to calibrate e2e before, now we can use this concept to calibrate vlead and lead acceleration too.
Verification
I have been driving a few days with another proof of concept branch and has been working really well on my GMC Sierra 1500 AT4 (2 inch taller than regular 1500's)
Route
Route: https://connect.comma.ai/fe80a4e1cedec853/1673461336764/1673461574279
As shown in the route, it maintained the distance with little to no overshooting or undershooting, also experience a cut-in that was handled perfectly with OP VOACC, I have driven a few hundred kms with this implemented so far