Enhanced lateral control

[Gernby]

This is a large set of enhancements for lateral control, which has been tested with positive results on many makes and models. The enhancements are in the following areas:

• MPC interpolation
• Projection of future state
• Steer rate and acceleration control
• Steering torque calculation (modeled as a resonant circuit)
• Longitudinal offset of EON camera

Overall benefit:

• Steering response is improved, and made consistent across all vehicle speed ranges
• Dependency and sensitivity to PID "tuning" is reduced
• Feed forward error due to suspension alignment, wind, and crowning is reduced
• Sample error from low resolution sensor data is reduced
• Steering mechanism is more direct / intentional, and has a more fluid sweep.

The proposed logic generally works well with standard PIF parameters. However, some vehicles, like the Honda Civic require some customization of SteerKpV and SteerKiV for smooth straight-line driving. To simplify the tuning process, I created another branch of the proposed logic, which captures high-resolution steering data to a CSV file for offline analysis. This would likely be done by community members with tuning expertise (via email).

The proposed tuning method uses a staged approach. The first stage will induce a sinusoidal adjustment of all 3 resonant steering parameters (resistance, reactance, and inductance) at different frequencies. The goal for this stage is to capture a high-level view of steering response from a broad set of combined values. This can be completed with only 10 minutes worth of data recorded at a steady speed on a well-groomed highway during favorable weather. If the steering response is not acceptable at any point in the 10 minute log, then SteerKpV and SteerKiV should be reduced by ~20% (or more) before repeating Stage 1. Otherwise, new values for all 3 resonant parameters will be chosen for the next tuning stage.

Here is a snapshot from stage 1. Note that the snapshot is interactive (after following the link). You can zoom in or out using the buttons at the top, or with mouse / keyboard shortcuts. Also note that the "Sway" value is a periodic agitation of the steering system, which helps with the tuning process.

The second tuning stage will use a fixed value for resistance, which should be the lowest acceptable value from stage 1 (ideally 0.5). However, the values for reactance and inductance will have a sinusoidal adjustment similar to stage 1, except with a smaller amplitude and frequency. The goal for this stage is to validate the adjustments from stage 1 while further refining values for reactance and inductance. This requires another 10 minute CSV data log similar to stage 1.

The third tuning stage will use fixed values for resistance and reactance. However, the inductance value will have a sinusoidal adjustment with a smaller amplitude and frequency than stage 2. The goal for this stage is to validate the adjustments from stage 2 while further refining inductance. This requires another 10 minute CSV data log similar to stage 2.

The fourth and fifth stages are optional, but allow the resistance or reactance value to be further refined with a sinusoidal adjustment while the other 2 parameters remain fixed. Stages 3-5 can be repeated again and again, if desired, until steering response is perfected (or "resonant", "impedence matched").

[icmma]

This has been work in progress since Nov of last year after 100s of tests can say these changes has been tested on almost every car as possible with help of other devs involved. This is a game changer as far as lateral control goes where @Gernby has put almost day and night to work on this awesome project in the end it was all worth it thanks @Gernby for your hard work bro:)

[Gernby]

FYI, I have added more content above.

[VirtuallyChris]

Gernby's work has made Toyota cars feel actually safe to drive with openpilot. Before I was getting severe jerking on the 2018 Rav4 Hybrid, but with a version of his steering I was able to tune and get smooth feeling turning - best a Toyota has ever felt! It also made my Civic perform better on roads with high crowns by staying dead center. Cars will need re-tuning, but I think it is worth it in the long run.

[rdm128]

I've driven for the last month on a fork of 0.5.8 w/ Gernby steering and I recently returned to stock for 0.5.9. There is a significant reduction in steering quality, especially in gradual turns where Gernby steering felt much smoother and more intentional.

[ghost]

Agree with @rdm128 Gernby's FF steering made my car feel like it was on a "track". perfectly centered at all times with a high degree of certainty that it would make turns very consistently. Although 0.5.9 made my wheel stiffer and centers better on crowned roads, it is a bit of a regression from what I am used to with Gernby's steering enhancement. The changes that Gernby's steering makes are subtle but noticeable when driving longer distances in curvy situations and in areas where there are high wind gusts.

[Gernby]

Thanks for the comments above!

I have done a successful merge of my code with 0.5.9, but still need to clean it up. Thankfully, it seems that 0.5.9 compliments the enhancements in this PR.

[Gernby]

FYI, I just finished testing a 0.5.9 version of my latcontrol, and will update the PR branch after I get more feedback from the community.

[geo2112]

Do you happen to know of a link that explains to a novice how to install it? I am worried it’s going to need SSH...

…

Sent from my iPhone

On Mar 4, 2019, at 1:46 PM, Gernby ***@***.***> wrote: FYI, I just finished testing a 0.5.9 version of my latcontrol, and will update the PR branch after I get more feedback from the community. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

[chassdesk]

geo2112 take a look at the workbench application on opc.ai will help you connect and install other forks ect.

…

On Mon, Mar 4, 2019 at 4:52 PM geo2112 ***@***.***> wrote: Do you happen to know of a link that explains to a novice how to install it? I am worried it’s going to need SSH... Sent from my iPhone > On Mar 4, 2019, at 1:46 PM, Gernby ***@***.***> wrote: > > FYI, I just finished testing a 0.5.9 version of my latcontrol, and will update the PR branch after I get more feedback from the community. > > — > You are receiving this because you are subscribed to this thread. > Reply to this email directly, view it on GitHub, or mute the thread. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#529 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/Ao9IrSzLNkMxlfxzWagwmhcdGv8DR3vLks5vTbG4gaJpZM4a8xJi> .

-- Thank You, Chas Setchell [email protected]

[Gernby]

After careful review of this combined set of enhancements, I believe some of the enhancements were unnecessary and / or problematic. I spent a considerable amount of time refining the remaining enhancements and will submit a new PR after feedback from the community.