Comma AI / OpenPilot Autopilot Progress on R1!

[Tatnai]

...or just run main fork open pilot until things are more stable on the other side. Works great, Rivian's longitudinal is better than dealing with instability.

Sponsored

 

[risingphoenix]

...or just run main fork open pilot until things are more stable on the other side. Works great, Rivian's longitudinal is better than dealing with instability.

Fair suggestion although the sunny pilot fork has all the main fork features with the addition of MADS, which is kind of a big deal.

 

[SoCal Rob]

I have some questions because I’m not sure I understand what you’re writing as you may be referring to human control or Comma control. I’m going to add my interpretation of what you wrote below each sentence so please correct me where I’m misinterpreting…

While it may be possible for the comma unit to function without engaging ACC (ie engaging the comma manually), I don't think that is ever the goal of Open Pilot.

While it may be possible to activate/engage the Comma without the driver using the stalk as if they were engaging ACC (by using the Comma touchscreen instead), that isn’t the way it’s generally intended to work.

I know of no other vehicle that comma works with that does not engage the ACC in the native vehicle way.

You know of no other vehicles that Comma works with where the Comma is not activated/engaged by the driver using the same control they used to activate/engage the vehicle’s native ACC before they added the Comma.

But in answer to your question, I am currently on a fork that does not utilize Rivian's ACC and those changes are still not allowed because the Rivian still "thinks" it's in ACC.

The fork you’re using doesn’t rely on the native ACC for longitudinal control but the act of engaging the Comma by using the stalk as if engaging the native ACC causes the Rivian to act as if the native ACC is engaged even though the native ACC is not performing longitudinal control.

Edit: The above is likely wrong because the comma already controls steering with MADS (without controlling speed). So it is already working (but not fully) without having ACC engaged.

All or some part of what you wrote may be invalidated if using MADS which allows for Comma-controlled steering while the driver controls the speed as they would when driving without driver assistance of any kind.

Please keep in mind that I have never seen a Comma AI unit in person so the entirety of my understanding is from discussion and videos. At this point I have one concern and my original question:

First, my concern relates to the Rivian’s native ACC which causes phantom braking events on curvy surface streets in my area very frequently and phantom braking events on highways far too often when there is a vehicle in an adjacent lane which would be in the vehicle’s path if the road was straight. I was hoping that the Rivian ACC would be completely removed from longitudinal control from OpenPilot, or at least one of the forks, to minimize or eliminate the phantom braking I’m experiencing.

Next, my original question about drive mode changes while Comma is engaged if it is providing longitudinal control. This seems to be an issue with the vehicle “thinking” native ACC is active because the Comma is typically engaged by using the drive stalk ACC action. I know almost nothing about the bus communications and what/where the Comma is reading/intercepting/modifying the data, but I just assumed that when activating the Comma with the drive stalk either 1) it was able to read the command to engage itself and strip it out so the vehicle did NOT get the ACC engage command at all or 2) the Comma would read the command, engage itself, and then issue a follow-up command to cancel ACC so the vehicle was not acting as if ACC was active. This follow-up cancel seemed easier and more likely to me.

Is it safe to assume that the Comma does neither of those things so the best case scenario is that the Comma is providing longitudinal control superior to the native ACC but the Rivian still behaves as if ACC is providing longitudinal control? I’m fine with that and will probably buy a Comma once longitudinal control is implemented and stable. The ability to change drive mode settings would be an added bonus.

Thank you for helping me understand this better!

edit: missed a comma (not Comma) which I added for clarity.

 

[risingphoenix]

I have some questions because I’m not sure I understand what you’re writing as you may be referring to human control or Comma control. I’m going to add my interpretation of what you wrote below each sentence so please correct me where I’m misinterpreting…

While it may be possible to activate/engage the Comma without the driver using the stalk as if they were engaging ACC (by using the Comma touchscreen instead), that isn’t the way it’s generally intended to work.


You know of no other vehicles that Comma works with where the Comma is not activated/engaged by the driver using the same control they used to activate/engage the vehicle’s native ACC before they added the Comma.


The fork you’re using doesn’t rely on the native ACC for longitudinal control but the act of engaging the Comma by using the stalk as if engaging the native ACC causes the Rivian to act as if the native ACC is engaged even though the native ACC is not performing longitudinal control.


All or some part of what you wrote may be invalidated if using MADS which allows for Comma-controlled steering while the driver controls the speed as they would when driving without driver assistance of any kind.

Please keep in mind that I have never seen a Comma AI unit in person so the entirety of my understanding is from discussion and videos. At this point I have one concern and my original question:

First, my concern relates to the Rivian’s native ACC which causes phantom braking events on curvy surface streets in my area very frequently and phantom braking events on highways far too often when there is a vehicle in an adjacent lane which would be in the vehicle’s path if the road was straight. I was hoping that the Rivian ACC would be completely removed from longitudinal control from OpenPilot, or at least one of the forks, to minimize or eliminate the phantom braking I’m experiencing.

Next, my original question about drive mode changes while Comma is engaged if it is providing longitudinal control. This seems to be an issue with the vehicle “thinking” native ACC is active because the Comma is typically engaged by using the drive stalk ACC action. I know almost nothing about the bus communications and what/where the Comma is reading/intercepting/modifying the data, but I just assumed that when activating the Comma with the drive stalk either 1) it was able to read the command to engage itself and strip it out so the vehicle did NOT get the ACC engage command at all or 2) the Comma would read the command, engage itself, and then issue a follow-up command to cancel ACC so the vehicle was not acting as if ACC was active. This follow-up cancel seemed easier and more likely to me.

Is it safe to assume that the Comma does neither of those things so the best case scenario is that the Comma is providing longitudinal control superior to the native ACC but the Rivian still behaves as if ACC is providing longitudinal control? I’m fine with that and will probably buy a Comma once longitudinal control is implemented and stable. The ability to change drive mode settings would be an added bonus.

Thank you for helping me understand this better!

edit: missed a comma (not Comma) which I added for clarity.

Yes, all of your interpretations of my poorly worded statements are as I intended them to mean. Remember, though, that I am merely an idiot consumer and do not speak for OpenPilot.

Regarding your concerns:
1) I agree that Rivian ACC (and even mores Driver+) is a poor implementation of ADAS. Even in its infancy, using the comma (specifically variants of the Sunnypilot fork) are in many ways superior to Rivian's ADAS and will only get better. While for me it is already better, in 1-2 months will be a very viable alternative for many Rivian owners who are unhappy with current ADAS (especially Gen 1).
2) The comma is a man in the middle but it still sends command signals to the AXM the allows it to control functional aspects of the vehicle, especially longitudinal (speed). For public use, I think the best implementation is going to be activation of ACC via the same mechanism it currently is activated (single press down on stalk). While in ACC (even if comma is intercepting and overriding), you cannot change drive modes. However, even currently, it is easy to engage MADS (again in the sunny pilot fork), which turns OFF ACC longitudinal control but maintains steering control. Then you can easily change drive mode and then reengage ACC with a single press down and continue your drive. Honestly, this is such an easy maneuver currently that I see no reason to overcomplicate the process and change it for this drive mode changing complaint. I am not aware of the specific safety protocols that OpenPilot has to follow but I'm sure they are pretty rigid and I doubt that there will ever be an option that "tricks" the vehicle into thinking ACC is not being used when it is by the comma merely for the benefit of being able to change drive modes.

TL/DR; the drive mode change is easy even with current sunny pilot and I personally see no reason to change it to make it different or easier.

To answer your query about when, longitudinal control with the comma has been implemented (for the last 24 hours or so lol) but is nowhere near stable. There is primarily one guy (Lukas) that is driving this development at breakneck speed and as long as he's engaged, there should be a much more publicly presentable option in 1-2 months.

 

[JM.]

Some commas are drawing too much power especially with forks that are doing more and you may need to install obd power to avoid this. In the meantime, if you have time I would advise plugging your comma into power in your house and installing sunnypilot/mads-tesla-prebuilt. It’s a bit more stable than Lukas new branch and will allow you to try MADS hopefully without the power glitches. Happy to help if you have additional problems.

Thanks, that's logical but also makes me wonder if now I've got a damaged harness, blown fuse...etc? I have Lukas' harness. The unit turned off, and would not turn on at all. It didn't give the low voltage warning you get on low power USB bricks. How would you branch out from the harness and also get OBD power?

There's an interesting discussion about how encryption may kill Openpilot in the long run, and this part was amusing. I completely agree, having worked in software/systems dev for nearly 30 years.

Anyway I have errands to do so I'll load the build you recommended, and go see if it turns on in the car. If not, I suspect something more serious has happened. Wonder if Lukas' harness has a fuse.

 

[JM.]

installing sunnypilot/mads-tesla-prebuilt.

Reminds me of a question...is the best/only way to change builds to factory reset the device? Minor pain to reconnect to the three wifi networks I use regularly, and do the whole tutorial, etc. The "uninstall" option just leads you to factory reset.

 

[risingphoenix]

Thanks, that's logical but also makes me wonder if now I've got a damaged harness, blown fuse...etc? I have Lukas' harness. The unit turned off, and would not turn on at all. It didn't give the low voltage warning you get on low power USB bricks. How would you branch out from the harness and also get OBD power?

There's an interesting discussion about how encryption may kill Openpilot in the long run, and this part was amusing. I completely agree, having worked in software/systems dev for nearly 30 years.

Anyway I have errands to do so I'll load the build you recommended, and go see if it turns on in the car. If not, I suspect something more serious has happened. Wonder if Lukas' harness has a fuse.

Others are having the same experience as you. For many there is a fix in one of the builds (should be included in the latest Mads-tesla-prebuilt I believe). There is an OBD power adapter that you can use but I also have the Lukas harness and don't currently have it. The idea of powering the comma inside via usb c cable is to eliminate a problem with the comma. It shouldn't matter from an installation perspective.

Also, I had similar problems initially and rotating the usb c cable as it plugs into the harness 180 degrees fixed my problems (must be a bad usb c cable). Easiest thing to try to start.

 

[risingphoenix]

Reminds me of a question...is the best/only way to change builds to factory reset the device? Minor pain to reconnect to the three wifi networks I use regularly, and do the whole tutorial, etc. The "uninstall" option just leads you to factory reset.

Yes you have to "uninstall" and then go through all the steps to install. You only have to type "sunnypilot/mads-tesla-prebuilt" into the Custom field though.

 

[JM.]

Thanks for the details, and I want to make sure one thing is clear. This was all working for weeks. Then simply turned off, and won't turn back on. So I really can't see how it's anything related to a bad cable or other pre-existing condition.

I have OBD power already for my phone charger. I am not picturing how I would add this to Lukas' harness but maybe I've forgotten/missed something. How would you physically connect both input power and the logic bus?

 

[godfodder0901]

Thanks for the details, and I want to make sure one thing is clear. This was all working for weeks. Then simply turned off, and won't turn back on. So I really can't see how it's anything related to a bad cable or other pre-existing condition.

I have OBD power already for my phone charger. I am not picturing how I would add this to Lukas' harness but maybe I've forgotten/missed something. How would you physically connect both input power and the logic bus?

The Comma harness comes with the propper adapter, called the Comma Power, and it should (may) be compatible with Lukas' harness as well.

https://comma.ai/shop/comma-power

 

[SoCal Rob]

Yes, all of your interpretations of my poorly worded statements are as I intended them to mean. Remember, though, that I am merely an idiot consumer and do not speak for OpenPilot.

Regarding your concerns:
1) I agree that Rivian ACC (and even mores Driver+) is a poor implementation of ADAS. Even in its infancy, using the comma (specifically variants of the Sunnypilot fork) are in many ways superior to Rivian's ADAS and will only get better. While for me it is already better, in 1-2 months will be a very viable alternative for many Rivian owners who are unhappy with current ADAS (especially Gen 1).
2) The comma is a man in the middle but it still sends command signals to the AXM the allows it to control functional aspects of the vehicle, especially longitudinal (speed). For public use, I think the best implementation is going to be activation of ACC via the same mechanism it currently is activated (single press down on stalk). While in ACC (even if comma is intercepting and overriding), you cannot change drive modes. However, even currently, it is easy to engage MADS (again in the sunny pilot fork), which turns OFF ACC longitudinal control but maintains steering control. Then you can easily change drive mode and then reengage ACC with a single press down and continue your drive. Honestly, this is such an easy maneuver currently that I see no reason to overcomplicate the process and change it for this drive mode changing complaint. I am not aware of the specific safety protocols that OpenPilot has to follow but I'm sure they are pretty rigid and I doubt that there will ever be an option that "tricks" the vehicle into thinking ACC is not being used when it is by the comma merely for the benefit of being able to change drive modes.

TL/DR; the drive mode change is easy even with current sunny pilot and I personally see no reason to change it to make it different or easier.

To answer your query about when, longitudinal control with the comma has been implemented (for the last 24 hours or so lol) but is nowhere near stable. There is primarily one guy (Lukas) that is driving this development at breakneck speed and as long as he's engaged, there should be a much more publicly presentable option in 1-2 months.

Thank you for helping me with all of this!

With the prospect of longitudinal control coming in the near future (1-2 months) I’m looking forward to joining the Comma community. I’ve read about Lukas and his work and how buying a harness from him benefits him so, assuming that’s still the case, I want to go that route.

The MADS to tweak suspension settings seems totally reasonable to me.

Regarding this, “… I doubt that there will ever be an option that "tricks" the vehicle into thinking ACC is not being used when it is by the comma merely for the benefit of being able to change drive modes.” I’m not sure I see huge benefit to letting the vehicle “think” it is controlling itself with native ACC but bypassing that and making the vehicle “think“ ACC is disengaged whole anything else (human driver or Comma) is controlling it. If anything, having the vehicle act as if ACC is in control when it is not seems more likely to result in strange behavior than having the vehicle act as if an external driver is controlling it.

 

[risingphoenix]

Thanks for the details, and I want to make sure one thing is clear. This was all working for weeks. Then simply turned off, and won't turn back on. So I really can't see how it's anything related to a bad cable or other pre-existing condition.

I have OBD power already for my phone charger. I am not picturing how I would add this to Lukas' harness but maybe I've forgotten/missed something. How would you physically connect both input power and the logic bus?

If you installed one of the new forks that takes over longitudinal, the increased power draw might be too much. Maybe go back and install base OP or whatever you used for weeks and see if it works again? Lukas has 2 versions of the harness and I have a version that doesn't have an easy way (that I can see) to get power. Once you troubleshoot you should reach out to Lukas on Discord (I know yuck) to get additional support.

 

[risingphoenix]

The Comma harness comes with the propper adapter, called the Comma Power, and it should be compatible with Lukas' harness as well.

https://comma.ai/shop/comma-power

Not everybody's harness comes with the power adapter. And Lukas has 2 versions and I think one is not compatible but not sure (I didn't receive a comma power with my Lukas harness).

 

[JM.]

I'm certain that power is not the issue, based on this... The device simply powered off. Past issues with power (using a crappy USB brick) would make it reboot. The device wouldn't power on with NO software, just factory reset. Then I rebooted the car, and it powered up. Then as soon as I put it in drive, the car spontaneously rebooted itself. The fun of open source/early testing.

I'm on the suggested Tesla build. MADS works fine. It's a compromise until there's a real solution.

@risingphoenix do you know if the "take over, excessive steering" is because of a hardware (car) limitation or the safety limits imposed in current builds? It's VERY limiting in my area.

 

[risingphoenix]

I'm certain that power is not the issue, based on this... The device simply powered off. Past issues with power (using a crappy USB brick) would make it reboot. The device wouldn't power on with NO software, just factory reset. Then I rebooted the car, and it powered up. Then as soon as I put it in drive, the car spontaneously rebooted itself. The fun of open source/early testing.

I'm on the suggested Tesla build. MADS works fine. It's a compromise until there's a real solution.

@risingphoenix do you know if the "take over, excessive steering" is because of a hardware (car) limitation or the safety limits imposed in current builds? It's VERY limiting in my area.

There are reports of this rebooting business that may or may not be related to the comma. I always do a Hazard reboot when I uninstall/install a new fork.

Yes, anytime you uninstall or install a new fork all of your torque settings are reset and it takes an hour or so of driving with it to recalibrate. So the good news is that the steering performance will improve. The bad news is that it still might be limiting for your area. If you can, upload a route and post it on the Discord for Gerry to parse your torque power. The data is suggesting that the lowest torque performance are OEM Pirellis and non-EV specific tires. So your specific tire can have a significant impact on performance as well. That's the bad news: different tires might handle curves much better. For now, MADS can be useful for these problematic curves as you can just feather the accelerator and let the comma continue to steer for you.

Sponsored