Having done a Seattle > SF > LA and back roadtrip with a Comma 2 back in March/April, this looks really cool.
Even with the very modest hardware of the Comma 2, it effectively takes all the stress out of highway / stay-in-lane driving. In my Audi A3, the car's ACC handles the accelerator and brakes, and openpilot handles the steering. It made the trip a whole lot more relaxing. It was also just a whole lot of fun to set up and tinker with, the community is great!
I haven't been able to observe anything close to the same quality of assistance system in any other car. Autopilot in a Model Y I drove just a few weeks ago felt plain erratic compared to openpilot, and the way it disengages completely when you make even just a small correction is really annoying. Audi's latest assistance systems silently give up as soon as the lane lines aren't 100% clear or things get a little too curvy.
If they can effectively make use of the additional camera data the Comma 3 offers in future openpilot releases, this will be well worth the price tag.
> Autopilot in a Model Y … the way it disengages completely when you make even just a small correction is really annoying.
Wait, so Autopilot disengages if you don’t apply enough torque to the steering wheel and disengages if you apply too much torque to the steering wheel? That doesn’t sound like an enjoyable autonomous driving experience.
Nope, it's really not enjoyable. To me it felt like fighting the car. I much prefer the "co-pilot" style of openpilot and most other lane assistance systems. Until you can trust your assistance system enough to not want to make minor corrections, Autopilot is always going to be an annoyance.
This is fundamentally where Comma and Tesla differ in approach. Comma's goal is to build the best possible level 2 system that makes driving relaxed and comfortable, and delivering good intermediates in the process. Tesla's goal is to have full self-driving, and any releases on the road to FSD are going to be a compromise.
I found Autopilot annoying until I got used to how it expects you to interact with it. At this point, it really isn’t too bad.
The nice thing is that a Tesla is always very clearly “under the control of Autopilot” or “under human control”. The downside is that it can be a bit frustrating figuring out how to convince Autopilot that you’re still paying attention (adjusting the the volume/cruise control, or briefly depressing the right stick is a lot less finicky than the torque sensor but you eventually get used to what the torque sensor wants.)
The pricetag is where I'm struggling to justify it right now. While the unit is compatible with Nissan's ProPilot it's not clear to me just how much more value I'd get out of adding the comma, but it doesn't seem like 10% of the car's value over what I already have.
Maybe with later software releases if it improves beyond 'the same as propilot but better at lane following'.
Reminds me of Carmack and Romero doing their thing with shareware games - true hackers, no fucks given, brilliant engineers, screw the corporate bullshit and get on with the business.
Speaking of sheer skills, take a look at geohotz twitch streams. Just the breadth of topics is fun to watch: making your own crypto, hacking the neural chip, low size encryption.
It is weird that the general opinion of Comma on HN appears to be positive while the general opinion of Tesla's Autopilot on here is generally more negative. It seems to me that all the complaints about Tesla including marketing, safety, and technological feasibility apply as much or more for Comma.
I agree that Comma deserves much of the same scrutiny, but it’s not that weird that HN gives them more credit than Tesla. Comma is the hacker, open-source, upstart player in self driving space.
Expectation management and effort to implement in your car as well. Buying a 7k "full self driving option" that's seamlessly integrated from the factory vs having to hack into your wiring to install a glorified phone.
I find this attitude towards Musk so, so odd. Where are you getting your info?
He was coding PC games as a teen, wrote the code for zip2, was the cto at x.com, and taught himself rocket science from textbooks (as confirmed by other rocket scientists). If that’s a “pseudo nerd”, who qualifies as a “real nerd”?
> "He was coding PC games as a teen, wrote the code for zip2, was the cto at x.com, and taught himself rocket science from textbooks (as confirmed by other rocket scientists). If that’s a “pseudo nerd”, who qualifies as a “real nerd”?"
If you don't code for 30 years, people might not call you a coder anymore.
Hotz on the other-hand still regular live-streams himself programming and hacking.
If anyone has to have the coding chops of Hotz to be called a nerd then we probably have only 50 nerds in the world. Y’all can definitely log off of HACKER NEWS since the news is definitely not for you!
I hate Musk as much as the regular guy but the constant undermining of the people you hate is the reason people who do this are never successful.
Consider for a moment That Musk is a Nerd, a ruthless businessman and a cock all at the same time. He made (or at least partly shaped) companies that definitely do fantastical things. If you keep underestimating him only you loose not him.
I'm always stumped by geohotz legend, he jailbroke the iphone on his own, then took some vuln from another team and used it on a sony device. Am i missing something else ? He seems super brilliant and fast but I think there's a lot of imaginary genius .. and in that way I don't think hotz differs much from Musk (overselling his dreams because most humans do that)
I think comma is mostly a joke, but at least it is only targeted at tech-minded who will probably put in the work to get it set up and understand that it is not foolproof. Compare that with someone just clicking a box on an order form and thinking they now have "Full Self-Driving"
The experience of learning about Comma, buying a dev kit, downloading a separate software package, flashing it to the device, wiring it up to your car, etc. (vs. just having it there on your Tesla, and everyone's heard of Tesla) is the source of massive selection bias.
If you're competent enough to get Comma running, you're probably technical enough (and invested enough!) to read the instructions and understand the operational limits of the system.
Yeah. We're approaching two million Teslas on the road now, and virtually everyone one of these has autopilot. I mean, yeah, with a sample size like that someone's going to find every edge case in manners ranging from hilarious to catastrophic. Just yesterday there was a guy on reddit claiming to have discovered one fork in Yosemite where, apparently, five Teslas have all wrecked their bumpers on the same rock. I believe it.
At the same time, it really is hitting all those edge cases. And the worst news in months is a cranky rock of doom?
I use this thing ever day. It's pretty amazing. I haven't driven Comma/OpenPilot yet, and wish them the best too. But the Tesla hate is approaching derangement at this point.
Tesla has over promised and under delivered so far. Comma has done the opposite. The Tesla beta software seems to be rapidly improving, so I've got my fingers crossed that Tesla pulls through in the end and manages to deliver something that lives up to the hype they have created themselves, but they have really dug their own hole at this point in time.
I once had a chance to go to George's house for a party, and all I can say is I was totally floored. I don't think I've ever been around a group of people who were so smart and just cool in my life. When I left a part of me really wished that I had somehow met a group of people like this in college or something. Just really smart, cool people, and most of all good. Everyone there was a genuine, mission driven person who couldn't have cared less about FANG type products. True hackers.
I once got a recruiting e-mail from comma.ai and the tone was very much acting like obviously I want to work for George Hotz and I'm so lucky he's willing to chat with me so what time should we schedule the meeting? I'd never heard of him nor the company and it all seemed very pretentious. They also sent it to my work address, which is not exactly classy.
To be fair, it was probably a bad recruiter, not George's fault, but that's now the memory that comes to mind whenever I heard comma.ai or George Hotz mentioned. Be warned, founders.
I'm in quite a number of different highly technical discords and none are quite like comma's. I mean that in the most scathing way possible unfortunately.
Disagreement and dissatisfaction are carefully curated and manicured to make it seem much more friendly, and put certain people in a much better light, than in reality. I've never seen so many posts deleted and users banned.
I know he dropped out and certainly I'm biased as an alumnus, but that's exactly the kind of people I met in college at CMU. I like to think he, like I did, got a little bit of that spirit during many late, late nights in Wean Hall.
I got to meet him for about 10 minutes at cmu while he was waiting for his photo shoot with wired magazine. He was a nice guy but also kind of a weirdo.
Nice guy? Weirdo? Yeah, that sounds like CMU alright. The friendliness of Pittsburgh will ensure the former; the school will exacerbate and exponentially develop the latter.
Plaid Parliament of Pwnage was a ton of fun too. I wished I had joined earlier and participated in more CTFs with those folks. I only actively participated in one, but it was such an awesome environment.
I happened to check today at what Tesla is driving for "Full Self-Driving Capability" and it's now $10,000. I was originally very interested in the feature, especially since it was available on a car that starts in the low 30s, but that price point seems way out of whack.
I'm just glad there's some competition that's not tied to the car manufacturers. I can't trust modern cars to have an 2014 iPad quality touch screen, I have no faith in their ability to deliver self-driving.
If you don't trust the car manufacturers, I'm not sure why you would trust a bolt-on system from a startup that may or may not have tested the system extensively with your specific car model. If anything, the car manufacturers at least have some legal liability.
They're much more transparent than any of the other car companies. There's also reasonable safety controls that I can understand. They don't apply torque or braking to a level such that I wouldn't be able to intervene. Essentially they can't run you off the road if you're paying attention. I have no clue what the other lane-assist guys do. They're a nameless faceless corporation. No identity, github, forum for issues, contact, nothing but marketing material.
They just released a $100/mo. subscription for the FSD features, but I think the more advanced FSD that supposedly can handle city streets has an additional feature flag that’s gated on your safety record or something.
Well done to them. One of the rare startups I've seen that are able to enter into this fierce assisted driving systems market without the hype, lies and deception with less funding and still profitable with hardware. That's how you do it.
This recent consumer report on comma.ai on the 'Overall Ratings Results' (page 6) looks very promising. [0]
EDIT: Care to elaborate as to what is the matter my comment about this hardware startup? Is what I just said about them all false?
I am surprised to see Comma Two in a consumer report not to mention out perform all other automative company including Tesla (probably the older version?). I remember in one of Geohot twitch stream bragging about their driver monitoring being really hard to duped.
Before watching the video, I was wondering what a "Comma Three" even is (you know, to decide if I should spend time watching the video).
So I went to their website first, to find out - and I think I still don't know. Doesn't describe the product on either the top page, nor the FAQ. Looks like a GPS/Nav device with AR features - but surely that's not worth more than two grand?
I have absolutely no idea if this is a good product or not (hell, I don't even need to drive where I live), but at least their website should tell me what is the selling point.
I thought the website was pretty good at explaining it? (I don't own one or know much about it other than having heard about it a few times on HN before):
> openpilot is open source software built to improve upon the existing driver assistance in most new cars on the road today. Tesla Autopilot like functionality for your Toyota, Honda, and other top brands.
> While engaged, openpilot includes camera based driver monitoring that works both day and night to alert the driver when their eyes are not on the road ahead.
It's basically an open-source (software) version of Tesla Autopilot and the Comma is their hardware they're selling that's already set up and running the OSS.
I agree that the website is very unclear for people who don't already know what comma do.
Their meta description is better, but still not great.
"Make driving chill. Upgrade your car with a comma three. comma is building the Android for cars. We have an open source driver assistance system that runs on most modern cars."
It's essentially an aftermarket version of the Tesla Autopilot system.
They have the same regulatory problem “dietary supplements” do. You can’t say something falsifiable like “makes you fall asleep in an hour” you have to say “improves sleep function.”
I bought the comma two about a month ago and really like it. Is there anything that comma three can do that the comma two can't do like lane changes without driver checking first, self navigation?
Not for another few months atleast.
The model they currently use on c3 was built using c2 data, but it still does perform little better because of better lens/camera on c3
Once they start getting data for c3, then they can really take advantage of c3 hardware including the wide fisheye forward facing lens by building models using that data. At launch, the model only runs on narrow cam.
C3 only features in future includes navigation like you pointed out and possibly stopping at red lights and stop signs. They might try to get redlights/stop sings on c2 but its unclear if they would able either because of camera fov or processing power.
lane changes without user confirmation might be possible on certain cars with blind spot monitoring but definitely not on cars without it. For confident lane changes you not only need to know if there is car in blind spot, but also speed of oncoming cars. This isnt possible without side radars or cams. So even the navigation part when it comes out will require user confirmation
comma three will be getting nav and e2e longitudinal while comma two will only be getting e2e longitudinal. the comma three will also get longer support.
I would never add a 3rd party device that controls my steering or other moving component of my car. Not even electrical including the main computer of my car. I get it, this is a new niche AI type stuff. My car is the last thing I will mess with hacker type devices like this.
The Silicon Valley bit is referenced several times in the linked video, including being a reason for the original company name.. and that this edition could lead to ",,," in annual sales (complete with an expensive yacht image).
>$2.2k for standard edition. same harness as comma two.
I'm still on an EON from 2018 so I'll need the new harness, which makes it $2400 for the base model. Really hard to justify the price of a loaded Macbook until they have navigation integrated with lane change and highway exits.
I'm honestly glad that they went all in and finally did a custom piece of hardware, but I think they could have cut $500 on the price by using an LCD screen and sold way more units.
Haha the fight between George’s ego and financial success is certainly a real thing.
But ultimately George is a hyper rational person and while it might seam like he is a no filter guy. He has a public persona and he uses it to sell the idea that they will never lower prices. If they need to lower the prices they will. That said I am not sure they need to lower the prices.
Even with the very modest hardware of the Comma 2, it effectively takes all the stress out of highway / stay-in-lane driving. In my Audi A3, the car's ACC handles the accelerator and brakes, and openpilot handles the steering. It made the trip a whole lot more relaxing. It was also just a whole lot of fun to set up and tinker with, the community is great!
I haven't been able to observe anything close to the same quality of assistance system in any other car. Autopilot in a Model Y I drove just a few weeks ago felt plain erratic compared to openpilot, and the way it disengages completely when you make even just a small correction is really annoying. Audi's latest assistance systems silently give up as soon as the lane lines aren't 100% clear or things get a little too curvy.
If they can effectively make use of the additional camera data the Comma 3 offers in future openpilot releases, this will be well worth the price tag.
Wait, so Autopilot disengages if you don’t apply enough torque to the steering wheel and disengages if you apply too much torque to the steering wheel? That doesn’t sound like an enjoyable autonomous driving experience.
This is fundamentally where Comma and Tesla differ in approach. Comma's goal is to build the best possible level 2 system that makes driving relaxed and comfortable, and delivering good intermediates in the process. Tesla's goal is to have full self-driving, and any releases on the road to FSD are going to be a compromise.
The nice thing is that a Tesla is always very clearly “under the control of Autopilot” or “under human control”. The downside is that it can be a bit frustrating figuring out how to convince Autopilot that you’re still paying attention (adjusting the the volume/cruise control, or briefly depressing the right stick is a lot less finicky than the torque sensor but you eventually get used to what the torque sensor wants.)
Maybe with later software releases if it improves beyond 'the same as propilot but better at lane following'.
A no frills, technology-first company that is uninterested in speculating their value.
I would bet any day on a company that doesn’t thrive by shilling, with conviction to win and sheer skills turned up to 11.
That’s the kind of folks that push civilization forward
Highly recommend - Masters of Doom audiobook: https://www.amazon.com/Masters-of-Doom-David-Kushner-audiobo...
Tesla overstates their capabilities and is more interested in keeping their stock price high than in doing what is best for consumers.
I find this attitude towards Musk so, so odd. Where are you getting your info?
He was coding PC games as a teen, wrote the code for zip2, was the cto at x.com, and taught himself rocket science from textbooks (as confirmed by other rocket scientists). If that’s a “pseudo nerd”, who qualifies as a “real nerd”?
If you don't code for 30 years, people might not call you a coder anymore.
Hotz on the other-hand still regular live-streams himself programming and hacking.
Consider for a moment That Musk is a Nerd, a ruthless businessman and a cock all at the same time. He made (or at least partly shaped) companies that definitely do fantastical things. If you keep underestimating him only you loose not him.
Just providing my opinion on why his reputation is different than someone like George Hotz.
If you're competent enough to get Comma running, you're probably technical enough (and invested enough!) to read the instructions and understand the operational limits of the system.
At the same time, it really is hitting all those edge cases. And the worst news in months is a cranky rock of doom?
I use this thing ever day. It's pretty amazing. I haven't driven Comma/OpenPilot yet, and wish them the best too. But the Tesla hate is approaching derangement at this point.
To be fair, it was probably a bad recruiter, not George's fault, but that's now the memory that comes to mind whenever I heard comma.ai or George Hotz mentioned. Be warned, founders.
I have had a Comma 2 for a while and the Discord has soured me to the company a bit. Lets leave it at that.
It's a refreshingly absurd, curious & blunt community.
Disagreement and dissatisfaction are carefully curated and manicured to make it seem much more friendly, and put certain people in a much better light, than in reality. I've never seen so many posts deleted and users banned.
I'm just glad there's some competition that's not tied to the car manufacturers. I can't trust modern cars to have an 2014 iPad quality touch screen, I have no faith in their ability to deliver self-driving.
https://mleverything.substack.com/p/commaai-flying-cars-and-...
https://www.tesla.com/support/full-self-driving-subscription...
This recent consumer report on comma.ai on the 'Overall Ratings Results' (page 6) looks very promising. [0]
EDIT: Care to elaborate as to what is the matter my comment about this hardware startup? Is what I just said about them all false?
[0] https://data.consumerreports.org/wp-content/uploads/2020/11/...
So I went to their website first, to find out - and I think I still don't know. Doesn't describe the product on either the top page, nor the FAQ. Looks like a GPS/Nav device with AR features - but surely that's not worth more than two grand?
I have absolutely no idea if this is a good product or not (hell, I don't even need to drive where I live), but at least their website should tell me what is the selling point.
> openpilot is open source software built to improve upon the existing driver assistance in most new cars on the road today. Tesla Autopilot like functionality for your Toyota, Honda, and other top brands.
> While engaged, openpilot includes camera based driver monitoring that works both day and night to alert the driver when their eyes are not on the road ahead.
It's basically an open-source (software) version of Tesla Autopilot and the Comma is their hardware they're selling that's already set up and running the OSS.
Their meta description is better, but still not great.
"Make driving chill. Upgrade your car with a comma three. comma is building the Android for cars. We have an open source driver assistance system that runs on most modern cars."
It's essentially an aftermarket version of the Tesla Autopilot system.
[1] https://www.youtube.com/watch?v=xzMUrB-Um1Y
But I think Geohot should not try to mimic some of the silicon valley tropes.
The communication is OK but sometimes borderline cringy…
Recommended trying out authenticity sometime, if you can manage negative emotions around fear/vulnerability.
Then you'll more likely recognize it in others.
As an aside, does anyone else get richard feynman vibes from the way geohot speaks?
I've definitely noticed it as well. It's that very specific northeastern accent combined with his sheer enthusiasm for everything.
C3 only features in future includes navigation like you pointed out and possibly stopping at red lights and stop signs. They might try to get redlights/stop sings on c2 but its unclear if they would able either because of camera fov or processing power.
lane changes without user confirmation might be possible on certain cars with blind spot monitoring but definitely not on cars without it. For confident lane changes you not only need to know if there is car in blind spot, but also speed of oncoming cars. This isnt possible without side radars or cams. So even the navigation part when it comes out will require user confirmation
The beginning of the video complete with awkward nerd dancing in a dayglo hoodie is also pitch-perfect Silicon Valley.
context: https://www.forbes.com/sites/ginapace/2019/11/29/diageo-laun...
I'm still on an EON from 2018 so I'll need the new harness, which makes it $2400 for the base model. Really hard to justify the price of a loaded Macbook until they have navigation integrated with lane change and highway exits.
I'm honestly glad that they went all in and finally did a custom piece of hardware, but I think they could have cut $500 on the price by using an LCD screen and sold way more units.
If you drive allot 2400 usd could be worth it. With that said I am not sure that sales will go down due to the price.
Normally I'd agree. But this is George's ego we're talking about. He'll never lower prices ever.
But ultimately George is a hyper rational person and while it might seam like he is a no filter guy. He has a public persona and he uses it to sell the idea that they will never lower prices. If they need to lower the prices they will. That said I am not sure they need to lower the prices.
$2200 plus $24 a month is a little steep.