Toyota: 2021+ RAV4 Prime

[jyoung8607]

Supersedes #1841.

• Message naming struct cleanup, similar to Chrysler later
• Consolidate alt flags if possible
• Add unit tests for new flags flags consolidated
• Check if any of the new messages have checksum that can be verified by the panda
• Ensure blocking 0x131 from camera doesn't affect any existing Toyotas (or just put behind flag) put behind flag

[sshane]

I would now label these three blocks with message names (ideally we make a struct that we can use instead of addresses soon)

[jyoung8607]

Do you mean trailing comments, like // STEERING_LTA_2.STEER_REQUEST?

[jyoung8607]

Added comments, I'm assuming that's what you were looking for if we're deferring an overall refactor of the entire safety to use message naming macros.

[sshane]

If we still use STEERING_LTA, do we no longer test that?

[jyoung8607]

We do send both. Both are present on the car, not sure why. Your point is correct. Reverted to Draft, I'll fix up the tests in the AM to verify no-actuation on both.

[sshane]

Did we ever check what the stock system uses? We were able to send LKA on some 2023 RAV4 but because the stock system used LTA for lane keeping (not just centering) we chose to switch off the unused API.

[pd0wm]

If I recall correctly the stock system uses the LTA for lane keep steering, but the torque message works fine too. So it's like most TSS2 cars in this regard.

They split up the LTA message into two parts, the old one which contains a bunch of values and flags (with no space for a MAC), and a new one that contains just the actual requested steering angle + a SecOC MAC. They did something similar for longitudinal.

[jyoung8607]

Now testing both messages, thanks for spotting that!

Verified by breaking the 0x191 check offline:

bool steer_control_enabled = lta_request; //|| lta_request2;

Which was properly picked up by both normal and SecOC tests:

FAILED test_toyota.py::TestToyotaSecOcSafety::test_lta_steer_cmd - AssertionError: False != True : req=0 req2=1 angle=0.0 torque_wind_down=0
FAILED test_toyota.py::TestToyotaStockLongitudinalTorque::test_lta_steer_cmd - AssertionError: False != True : req=0 req2=1 angle=0.0 torque_wind_down=0
FAILED test_toyota.py::TestToyotaSafetyTorque::test_lta_steer_cmd - AssertionError: False != True : req=0 req2=1 angle=0.0 torque_wind_down=0

[sshane]

@adeebshihadeh do you want to merge this with LKA when we specifically didn't want to merge LKA for the new RAV4s which don't use it but could? commaai/openpilot#29978 for reference.

[jyoung8607]

It's probably possible to switch to LTA, the SecOC part should just carry over. However, it will take multiple round-trips with test owners to plot how the stock camera drives the two LTA messages in tandem and then replicate, test, and validate safety for that behavior.

I don't have a preference on whether to make the switch, the comma openpilot team knows more depth and breadth of Toyota history than I do. If we do want to make the switch, my preference is to ship what we have and create an issue to track migrating to LTA.

[pd0wm]

If I recall correctly the stock system uses the LTA for lane keep steering, but the torque message works fine too. So it's like most TSS2 cars in this regard.

Sorry, forgot lane keep is what you should call the safety nudges to put you back in the lane, not lane centering. I did not check what the stock system uses for LKA.

[jyoung8607]

Verified the stock system actuates on both STEERING_LKA (torque) and STEERING_LTA_2 (angle) depending on conditions.

[jyoung8607]

Just a note for review: I originally wrote this to build lta_angle, like the other LTA message. Unfortunately the mutation test doesn't like it. We never have actuation via this path, so the angle is never valid while non-zero, so the mutation test sees it can introduce faults in the bit-shift operation without consequence.

[sshane]

Looks much cleaner!