WATCH UPDATED VIDEO COMPARISON ➡️ th-cam.com/video/Hc8G5U0p1sA/w-d-xo.html 🚘 TESLA RESOURCES Best Model 3 accessories: geni.us/tm3accs Rent a Tesla: geni.us/Turo Buying a Tesla? Free Supercharging: geni.us/t3sla Geeky Tesla shirts & prints: geni.us/SFSF My best Model 3 videos: geni.us/cjUF9g ⭐️ FAVORITE RESOURCES My TH-cam camera gear: geni.us/YTgear TH-cam Success course: geni.us/YTcourse Real estate investing: geni.us/Fundrise Stock investing app: geni.us/qZkq Best savings account: geni.us/Wealthfront My Amazon store: geni.us/AMZN 👇 JOIN ME TH-cam: geni.us/AndySlyeYT Twitter: twitter.com/slye Instagram: instagram.com/slye Email squad: geni.us/TeslaFan
Andy Slye great idea to put this information as a post vs just in the description that I never really take the time to go open and look at .. it’s a PITA on some device..
Andy - Open pilot lacks the Basic Safety systems found in any Tesla with AP3. also, 8 cameras and RADAR and SONAR trumps just one camera. Open Pilot cannot use ANY of the Built in Cameras in your car. TESLA can Automatically detect cars in FRONT, Front Left, Front Right, Both sides, Both Bind Spots, Rear and Rear Left , and rear right. Open Pilot cannot warn you that you are About to be Rear Ended. the TESLA RADAR can Actually see the Road Ahead even Underneath cars in FRONT of you looking for Road Hazards. Tesla AP3 is Level 3 - 4 , its why its called Autopilot 3. Tesla AP3 will be Improved over time and TESLA is Already working on AP4 chips. and Open pilot does NOT have Billions of miles Driven. TESLA AP3 FSD is the Worlds most Advanced Autopilot. Open pilot cannot slow down the car to a stop if you have a Heart Attack or other medical situation. Lets see how well Open Pilot does on a 5,000 mile trip. that test is EASY for a Tesla. Lets see how well Open Pilot does in DUST, FOG, or SMOKE ? RADAR negates the lack of vision, allowing you to FLEE in a Natural Disaster.
@@markplott4820 I am a TM3 driver but I have to correct you. First there are fours versions of Tesla autopilot: HW 1, 2, 2.5 and 3. It is not called HW 3 because it is level 3 and indeed it is not. It is currently level 2 and it will be level 4. HW 1 is using the MobileEye computer and only has one 1 camera. HW 2 is Nvidia with one GPU and 8 cameras. HW 2.5 adds an extra Nvidia GPU for redundancy. HW 3 swaps out the Nvidia GPU for a Tesla made neural network processor. HW 2 and 2.5 can be upgraded to HW 3. HW 1 can not be upgraded. Three cameras all look straight ahead. The difference is the angle of field of view (one camera looks far ahead in narrow field of view, another has a very wide field of view and the last somewhere in the middle). Two cameras on each side of the car, with one camera looking forward and to the side, the other rear and to the side. One camera at the back of the car which is NOT used by autopilot. That one is currently only to help park the car. Autopilot does not currently use all these camera to their full extend. It is mostly using one forward facing camera, which is why it also sometimes has trouble with very tight curves. It simply can not see the road even though it should be able to, if it would just switch to a side mounted camera or the very wide field of view front camera. It does use the side mounted cameras to detect traffic during lane change however. Think about it like this. For each camera Tesla trained one or more neural networks. The one running on the front camera is tracking lines used to center the car in the lanes. The neural network running the side cameras however were not trained for lines. Instead they trained that to track cars and objects, such as traffic cones. And by the way, autopilot disengages during bad weather. You can not even use the radar to get old fashioned cruise control. Technically that would be possible but Tesla does not allow it.
@@BaldurNorddahl - All current Model S/X/3 are AP3 computers since April 2019. HW 2.5 since September 2018 can get Updated hardware in the AP3 chips. Both HW1 and HW2 are Legacy systems and cannot get Hardware Upgrades but the can still get Software updates. AP3 is Level 3 - 4 Period, it can Navigate on streets, to the on Ramp, Speed up + change lanes, pass SLOW moving cars, Navigate to Exit, take off Ramp and drop you off at work, this is possible with AP3 v10. this latest Software update is much improved since AP3 first came out, v10 is a game changer. a Friend of mine Moshe the Electric Israeli recently completed a 3,000 mile Road Trip completely using Autopilot. and I am sure AP3 v10 can also do a 5,000 mile Trip.
Just realized self driving will put a stop to a lot of ticket revenue gathering. What will they do to get their money after that? The same thing? Will they get fired? Hmm.
@John Buick It's referring to the big argument going on right now around with Jordan Peterson... It's just a joke if you follow the news on the gender names battle. It's been going on for min 2 years.
This is probably the reason that George Hotz had his peepee smacked by the NHTSA a few years ago. I've noticed that this software seems to be limited to cars that already have sensors installed. For this to be a truly revolutionary thing, it would have to work with aftermarket sensors and include a suite of calibration tools ... and possibly even require professional installation. Simply strapping a cell phone to your ECM and pointing your 2-ton missile down the highway while declaring "IMMA ROBOT NAO" is not sufficient safety testing.
@@chooseymomschoose "it would have to work with aftermarket sensors and include a suite of calibration tools" it is open source, whats stopping you or anyone else from coding that feature! I predict that the rise in demand for self-driving cars, and the fact overtime, features tend to cost less, it'll be more common to see these new cars ship with enhanced driving capabilities, allowing a larger group of people to take advantage of commai
OK that's the downside of this as far as I can tell atm. My car allready has a good ammount of features and sensores that make is almost capable of self driving. I have lane assist, blis, forward assist, cruise control, camaras in the front and back and all of that stuff. Basically all it would need is a software that brings the informations and features together so it would drive fully autonomous. I mean, with cruise control and lane assist enabled it allready feels like autopilot the problem is that lane assist is designed to only work and take control if you're going to cross a line instead of "fixing" the car in between the lanes ...
@1234 Exactly. From my perspective, you don't even need extra censors, add rear view camera(s) and you're good. Humans already are only driving with their eyes, as long as the system can see the cars, it would still react much faster than we can and provide way better results. All these fancy sensors in Tesla's seem to be a more reassurance and increase price than necessity.
The same way that you don't know about wayve, the system being used in the Jaguar i Pace. It uses machine learning. The original was developed using the Renault Twizy. th-cam.com/channels/NERNUuB5kAj7rGGfhk0C3A.html
@@aslye very impressive to see it and thanks for showing what is possible, BUT there are still big problems and not at all impressive how you drive or let the car drive and run "dark-orange" or red lights (which we jokingly call "cherry green" :-) about the "open system": there is already quite some discussion about the responsibility on "official" systems, and they are equipped a lot better with many more cameras, sensors, radar, lidar, etc. having such systems on many "less suited" cars will increase risks, and might make it more difficult to increase acceptance of any(!) kind of self driving in the public opinion and in changing laws to make them legal. afaik, at least in europe, such "private modifications" currently are not legal and invalidate any permission to use those cars in public traffic, as well as invalidate the mandatory insurance. in addition, if it is really "open source" and everybody can modify it and make his own rules and determine the degree of aggressiveness, it will be turning even worse and not make selfdriving cars safer in general. about the traffic lights: on the widening lanes, you are so concentrated on watching how the car handles the widening lanes that you forget to check the traffic lights. several times the light switches to yellow at least three seconds before you reach the crossing, eg at 6:42 the lights turn yellow and when you enter the crossing at 6:45, the lights just turned red ! i don't live in the usa and maybe you still may cross roads like that, but where i live you have to stop in front of the crossing when you safely (depending on speed and traffic behind you) can break. three seconds should be "good enough" to do that, and on your red, other cars might already try doing a jumpstart from the sides before you finished crossing 4-5 seconds after first seeing "yellow" ...
@@Anson_AKB those systems rely on also not braking unaturally fast, because then the programer would get blamed for getting rear ended. just as humans, AI can get blinded or miscalculate LOL
@@Anson_AKB I encourage you to watch some talks from George hotz. You make a lot of false assumptions in this comment and I think it would benefit you to learn more about adas systems and what is possible with current technology.
Excellent side by side guys. I am totally impressed with OPEN PILOT, but, as a rule, I never use cruise ctrl on local streets like those you traveled on anyway. So, I am totally blown away that both system even worked at all. This was soooo Eye-opening and informative straight up. Phenomenal work guys, and please do more analysis like these in the future.
+Steven Smith Here in the UK the roads are really busy so often people don't maintain the distance they should... It's quite common to see a Tesla with its blinker on to change lanes, then time out, then the blinker is back on, then it times out again. At peak times the Tesla will never find a gap big enough, whereas a human driver might take a greater degree of risk by pulling into the smaller gap.
Most ADAS systems use a single camera + radar. You could get away with a mono camera, with less redundancy, but then again, comma.ai isn't exactly going to get in hot water with the NHTSA.
My two reactions: (1) Paul Rudd is bloody everywhere these days! and (2) the one thing Openpilot has over Tesla is you can take it with you to the next vehicle.
TESLA USES REAL WORLD INTERACTIONS, WHICH ARE FAR SUPERIOR THAN THE COMPUTER SIMULATIONS WHERE INPUT MUST BE ADDED. THERE ARE WAY TOO MANY WIERD THINGS THAT GO ON LIKE A FULL WHEEL TRAVELING LIKE 30 MPH OR SO HEADING STRAIGHT AT YOU ON A HIGHWAY.
I feel like I would be much more comfortable having the view the Tesla has because it shows that the computer is detecting everything around you where the other system you just have to trust that it sees things but there is really no way to verify.
It's hard to tell in this footage but openpilot shows the detected lane lines and path as well as the lead car being detected on the screen mounted on the windshield.
Teslas view... well...it doesn't show what the computer detects. It shows what the computer THINKS it has detected, and nothing more. And that is a problem, especially since YOU as a driver should have your eyes on the road at all times, and NOT be looking down on a screen. It makes it Less safe to have a lot of distractions in the car.
@@ItsaB3AR Yes it is but: It's placed so that you do not have to move your eye so much and, it doesn't show a lot of fancy icons (cars, trash bin(!?), cones, ...). And, OpenPilot watches you, so if you look away too long, it will nag you about it.
@@ryanDogAdventure Toyota did have some good features in the gen3 with advanced tech package. The system in the gen4 and Prime got worse and is basically useless. Openpilot is a much needed upgrade.
@@techie163 that must be outside of Japan. As of last year, almost all brands, Subaru, Mitsubishi and such have level 2 systems in them. I think Japan is gearing up to making it mandatory for cars due to seniors causing accidents. They are really focusing a lot on this here in Japan.
@@pimpstick2 LOL. That's not the way neural networks work man... You think open pilots hardware (which is basically a phone) is running a neural network? No, its not. Lounge lizard is correct that you would need faster processor when using open pilot on your car if there were extra camera angles. The neural network is what compiles data (video) from every car, mashes them together and compiles a driving model or software update that then goes out and is sent out to all the phones to run. Consumer hardware aren't running the neural networks. Currently open pilot will never be able to evolve past level 2 autonomous driving unless in the future they are able to add extra hardware and cameras to the sides and rear of the car. They currently have no plans to do this yet and are mostly just focusing on perfecting L2.
As a Tesla owner, I think OpenPilot is great. Not that it’s more capable, but that the more cars that have some kind of drivers assist capability, the better. Unfortunately, the one ICE vehicle we still have, a Chrysler Town & Country minivan is not on the compatible list yet. If it does become compatible, I would add OpenPilot in a heartbeat.
Unless your Chrysler minivan has a motorized steering rack, it unfortunately will never be compatible. OpenPilot can only support vehicles whose steering rack is motorized.
Musk should have hired him when he had the chance. Of course there would have been personality conflicts that would have ended the relationship pretty fast. Hotz would have advanced Tesla's vision system in months, rather than the years it is taking.
ram 1901 I definitely think he could have been as asset, but I agree I could see their personalities clashing lol. And I’m not sure he would revolutionize what Tesla is doing as he admitted himself that Tesla will win FSD in that interview. Far more visualization tech is ready at Tesla but they need approval before each new release by regulators
Those roads are in very good shape. Lack of lanes and patches for humans is not a problem if the autopilot is incapable of driving in such conditions than it is definitely not good enough to be adopted.
So awesome that both of you fairly and honestly assess the two products. It would be great to have some competition between the two so that both can up their games.
This was absolutely fascinating! Thank you guys for taking the time to put Tesla Autopilot and Openpilot through their paces. I had no idea there was an self-driving aftermarket option like Openpilot; I'm impressed with how well it performed! It will definitely factor into my next purchase when it comes time to get a newer car. Please do more of these! Would love to see how they compare in situations like heavy traffic, busy parking lots, even dirt roads!
Amazing comparison! I've been using openpilot for years, even before they sold any products. It's been amazing to watch my Honda Civic get better and better with ever openpilot update. I almost never manually drive during road trips now!
I think AutoPilot is functionally better, but OpenPilot is more technically impressive due to the insane simplicity of the hardware yet it still copies a lot of AutoPilot's functionality.
I was also surprised it did this well. I'm just speculating, but I wonder whether the limited data (just a single camera) makes it actually easier to train those NNs for the basic cases, but over time it will start losing behind Tesla, which will be able to evaluate more complex situations better.
@@peter.g6 You're probably right about that. Right now, it's easier to catch up, but it will eventually lose out. Even though Tesla's system is more complex and more difficult to train, it's also, well, precisely that - more complex. That means it will be able to better deal with more complex situations once it takes the time to be trained.
I'm not sure it sopies Tesla. Openpilot has been around since 2016 now (and earlier as prototype by the developer), and they take it their way. It's not about copying something - it's about getting it to work.
What a great video. I use OpenPilot on my Volt and it's mind blowing how good it is for being a fairly low cost cell phone in a 3D printed case. Comma has done some incredible work. For the follow distance issue: Since a a large amount of OpenPilot is open source there are a lot of community forks that add features. I use Kegman's fork which has 4 distance settings. Makes a huge difference. I can also disable the steering and just keep speed adjustment enabled. This is really useful for construction zones or weird roads with bad markings. I can take over steering and let OpenPilot handle speed. This all integrates nicely into the car. It uses the car's built in LKAS button to control whether steering is enabled or not. I am jealous of his Toyota's steering torque though! My Volt needs a bit more help on sharp curves especially at lower speeds.
@@Flankerr The cell phone component was $500, not $1,000. I could have built it for cheaper, but I didn't feel like messing with that. The total cost of the system, including custom GM stuff, was $1,000. I also could have built the GM wiring for cheaper, but again, it wasn't really worth my time. I would have gladly paid Comma $1,000 for just the cell phone portion of it. I get enough utility from it and it's worth supporting a small, innovative company. The driving model is not open source which is what most of their research goes in to. It should also be noted that this isn't a real product yet. They're selling dev kits to help fund the cost of development and get community support. These should not be purchased by regular people yet. They require a fair bit of tinkering to get working well. They'll probably come out with a product eventually that's much easier to get running.
@@Flankerr: Damn, Tesla Autopilot is many times that expensive and in the same class, re function. After ALL this time, false promised by Musk, hardware upgrades, etc.
Great video. I've been curious about a comparison of the two system. George Hotz the Founder of Open Pilot seems very dedicated and passionate to bringing this system to as many drivers as possible. Once he and his team succeed in perfecting the system. I'm sure they will license it to traditional car manufacturers.
George Hotz is a genious. I watched alot of his livestreams and he said whenever Tesla solves level 5 autonomy comma ai will solve it too after 2-3 years. No doubt though that Tesla will be the first of all the companies trying to get to level 5.
Not sure if this is still the business plan but GeoHot recently said this on Discord: "We’ll leave the hw game in 3-5 years when Xiaomi and Huawei are making Openpilot devices (based on Qualcomm or NVIDIA chipsets) There will be plenty of 3rd parties making hardware by then, some with our blessing and support, and they'll be able to use trademarks like "openpilot" and our logo ... such devices will be allowed on our network and will be supported with upstream OP also. OP would be just a software update ... we want to control the OTA on cars and use the OEM’s compute and have a monthly subscription business model. There will also be devices without out blessing, branded differently, not on our network and requiring you to run forks, exactly the same as Android. Still gonna be 5-10 years before OEMs get the picture and include it 1st party." Source: i.imgur.com/GKdv3IH.png
Honestly i like how careful openpilot is especially during heavy traffic. I cant wait for self driving to become a thing.. I cant drive for medical reasons (NEVER have been able to and im 32 now) so my life has been SUPER limited. can't get jobs i want because i have to work within 5-10 miles of work or else im biking longer than im working. I cant go get groceries easily (though now people deliver so thats cool). i cant even hang out with friends really without paying money for a taxi or relying on a friend.
Wow hold on here... Open pilot never plans to be automonous and is meant as a simple lane scanner to follow said lanes. It has one to none safety features while autopilot has a group of cameras giving you 360* coverage and full automonous driving in the future allowing people who can't drive have the computer drive for you
@16:45. Actually having that much of a gap between cars is ideal given your speed. It's this mentality of not leaving long enough gaps that cause traffic and accidents.
Yes. The driver is tailgating. Tailgating is dangerous, especially on highways: the braking distance (and kinetic energy of a potential crash) both increase with the SQUARE of the speed.
Great comparison. I’m so geeked you drove literally to my workplace. I take the same route to work every morning and I was so confused on why everything looked so familiar to work, until I realized you really drove through where I work 😂
Thanks for doing this comparison video, Andy. I heard George Hotz speak at our company's conference in December 2018 and met him personally. Comma AI's stuff is really great and I hope more car companies will start partnering with them. Just couple of quick things I wanted to point out: 1) Tesla's following distance isn't car lengths but it's "time-based" (not seconds, either). 2) Tesla doesn't recommend anyone using AP on city streets or windy roads (they state this in the manual), that will be available when FSD is developed and Tesla is rolling it out. Excellent video as always!
Model 3 specifically states in the manual that it isn't measuring in terms of cars but rather seconds. When you set it to "2" it isn't two car lengths, but 2 seconds of space. Check out the manual, someone showed it to me when I made the same mistake!
As a Tesla owner, I think Open Pilot looks excellent. We need as many cars as possible to be on systems like this to reduce accident rates. Having an affordable system that can be added to any car seems a great step in the right direction.
Civics have waay less steering torque compared to this video but it has enough for highway driving. Stay tuned though because civic firmware hacks that increase torque and bring steering to zero are going to be released soon if you're adventurous enough.
@@edwin260 oh I already know the steering torque on the Civic is garbage lol. I've seen a few people pull off firmware hacks for that motor though, is there any dev that planned on releasing to the public?
@@1blackbirrrd Check out the official comma ai discord. They recently added a firmware modding channel if you want check on the progress. The devs (not comma employees) briefly released a guide on how to do it with the necessary scripts on github, but they decided to take it down in order to refine it. But public release definitely coming soon.
The only thing I didn't like about OpenPilot(owned for 2 weeks prior to buying Tesla) was the fact that you couldn't leave the ION Phone/Unit in the car due to heat issues. And your compatible car would only be as good as it's assistive cruise and lane keep assist capabilities.
This is a really big deal! Let's look at how this terrain could work with Tesla: 1) We will need entry-level products as a migration points. Many people are up for spending $1k to test out a technology. If Tesla's is vastly superior then after confirming "the concept is good but I'd like better" something like open-pilot becomes a bridge for Tesla buyers to want and value superior autonomy. 2) If Tesla will partner/produce a $1.5k version like this that is as good or better than open pilot - or a partnership, there is a huge win/win: a) Safety for drivers should improve, saving lives, building Tesla's brand. b) Tesla can gather more miles driven data and learn how to make it's main model more efficient and measure the safety discrepancy per million miles, leading to many contributions to perceived values. c) By far, Telsa's largest asset is owning and selling/updating the safest most versatile autonomy software in the world. Most of the value will come from subscription services to as many cars as possible on the road. Tesla can sell design-config and software support to private drivers and legacy auto-makers as a way of keeping customers. This is pure profit on Tesla's bottom line. 3) If Elon does not embrace non-Tesla partnerships and open-source, a partnership of the main legacy automakers, venture investors and home-kits like this probably offer the biggest challenge to Tesla over time. The autonomy piece is huge, and much like android took over apple in market-share because it collaborated rather than controlled, the collaborative open source block-chain model of sharing data is the ultimate winner in all spaces. 4) I'm excited to try and fit my tundra with one of these set ups, which I'm sure will become more and more versatile over time. Thanks for telling me that they exist! I'm happy to spend more for better safety so I really hope that this becomes a huge industry: a) People making videos on the safest, tested set-up for every vehicle ever made (For the Tundra 2001 you want five cameras from this source mounted exactly here in this way, then you want the screen attached here, and a subscription with X company and a record set-up with X hard-drive.). Once this get's proven (Tesla currently has 8x safety factor over humans which is enough to get it but not enough to get a new car if you don't have a Tesla and like your cars features) then literally everyone will retrofit existing cars and all new cars will have these features. The exponential tech-price curve dictates that every few years something twice as good will be available for the same price and the profit margins are so high that competitors will enter the market until you can buy safety per mile for almost nothing and we stabilize at 7-12 times safer than good drivers globally. Tesla has the lead, but eventually all companies will crack this one way or another very cheaply because of the exponential learning of A.I., hacking, intellectual property theft, better simulations and time. It will be a commodity with few distinctions in twenty years for a few cents a mile to have top-line A.I. guiding all transportation including planes, boats etc. because it will save money and be safer. Andy, I appreciate you showcasing and giving points to a cheaper competitive model given your passion for Tesla. That makes you trustworthy as an information source. Sincerely, Dane
"Off-ramp slow downs" are actually based on the speed that other Teslas drove around that same corner, so autopilot references what other Teslas have done and uses that to set the speed on that same corner. From page 87 in the Model S manual: "Off-ramp speed: When enabled while on a highway interchange or off-ramp, Traffic-Aware Cruise Control may reduce your set speed ...to as slow as 25 mph (40 km/h) to better match the reported speeds of other Tesla vehicles that have driven at that specific location."
This is one of the best side-by-side comparisons of autopilot systems I have seen, with two drivers and two vehicles following the same route. There's a very good step-by-step replay and commentary about how each system handled various challenges. I am really surprised how a single-camera, smartphone-based system can compete so well with a much more layered technology in the Tesla.
Great comparison between AP and OP! My wife did a presentation at PepsiCo a while ago about AI Machine Learning and she did mention both Comma.ai and Tesla. It was my first time hearing about Comma for quite a while and am happy to see how well they've been progressing!
Thanks for the vid. I have Garmin GPS that tries to warn of lane departure and it gets confused with tar strips as well. It also does collision detection and notifies when the car in front moves (for stoplight texters). Single camera, it is pretty impressive. It does speed notifications and traffic. You can also add another camera via bluetooth. It would be cool to see the GPS technology married with the opensource stuff.
I didn´t even know Openpilot exists. I am positively impressed by this device. It deserves all respect for its functions and affordable price. Thanks for the great video!
Good video. Surprised to see openpilot performing just as good or better than Tesla in some situations. I considered it when I had the Chevy Volt, but I'm not really a DIY type of person when it comes to my car.
Btw the blue steeling wheel icon means auto steeling. The blue circle around the speed means the advanced adaptive cruise. These two together is the basic autopilot now.
OpenPilot is amazing. I mean autopilot is the superior hardware/software here but then again it has loads of cameras and native support from the ground up. OpenPilot is a fraction of the cost and basically plug and play for compatible cars. Amazing tech with very, very decent performance.
Nice to see both systems side by side. Openpilot is impressive. Wonder how Mobile Eye does compare those days. The more competition the better. Thanks for your videos.
MobilEye is far beyond Tesla's Apr. 2019 FSD demo, but they did start ~10 years earlier; see this recent 25min drive through Jerusalem: th-cam.com/video/hCWL0XF_f8Y/w-d-xo.html
@@thisisntsergio1352 Yeah, but the Corolla has the capability of using OpenAI? That means the Corolla has servo-motors connected to the steering, the braking, & the gas! Corolla is one of Toyota's lower end cars. I'd say down from that is a Yaris, so Corolla would be in the bottom two. When did *that* happen? I guess about the same time dogs learned to talk & people learned to fly unassisted. I drove a Camry that had lane assist, and that was awful. Turning on lane assist was more akin to bowling where they put those long pads in the gutters so kids can't throw gutter balls: I ping-ponged back and forth in the lane. Automatic driving isn't like that. I like the idea of OpenAI. I run Linux & use opensource software all the time. But having that capability means all the other pieces of the self-driving puzzle have been put in place, which astounds me. I used to have a Corolla. I drive Tesla (& others), and find it hard to imagine a Corolla can compete. Plus, Corollas aren't electric, and as much as I wish it weren't, being all electric is a huge deal.
@@phillipcrittendon9172 I drove a Camry that had lane-assist and didn't like it at all. I know a lot about Toyota. Way more than 99.9% of the people out there (I provided critical consulting to their US operations for years) and I think they're in trouble.
The fact that most have no idea that other manufcaturers have similar competitive systems shows the value of Tesla's PR hype machine . It is, if not accurate or truthful - at least supremely effective.
@Andy Slye , you keep saying Tesla Autopilot (TACC) lets you set car lengths as the following distance but they absolutely are not 'car lengths'. The AP (TACC) setting of '4' is not the same distance at 70 mph as when you are traveling 35 mph. It is 'time-based'. Tesla manual: "Each setting corresponds to a time-based distance that represents how long it takes for Model 3, from its current location, to reach the location of the rear bumper of the vehicle ahead of you."
How can a informative video like this gets dislikes? If u dont like andy just dont follow him wtf, great job, I wish I could work in a software company that programs autopilots
That's awesome!! Andy and Logan, you both should do 2.0 comparison video for Tesla FSD Beta and Openpilot 0.8. Both had their upgrades, so it would be worth seeing the improvements.
I've found that Teslas won't auto-lane-change across a solid white line, and has trouble with changing lanes with very tight dashed lines. In earlier updates, say maybe in November 2019, if you started an auto-lane-change when the lane lines were dashed but you started entering a solid white line section, such as in construction zones where they shift the 2-4 lanes to the left or right, the car would jump back to the other lane. Now, at least in 40.50.7 and a bit earlier, if the car has already started changing lanes, even if the line turns solid the car will complete the lane change. I'm thrilled to see how far comma.ai OpenPilot has come! It looks like Autopilot wins in more conditions and offers a bit more of a smooth experience in most conditions, but it's great to see where OpenPilot beats Autopilot too. Great video!
Honestly I went into this thinking Autopilot would beat Openpilot hands down, no ifs ands or buts, but using 1/8th the cameras, no radar or anything else this was very impressive! It will take a long time for them to get the miles Tesla has right now, but it's surprisingly effective for what it is!
Man, this video kind of show just how ready, but also not ready autopilot is. Can't help but to be at awe how the human brain can navigate these courses without even really thinking about it. It's so amazing! Autopilot doesn't have to eat sleep or take a break. Its whole existence and design were meant to navigate these streets. and it can't even do it as we can. It's going to be uncanny when it finally can beat us by machine learning!
Thank you for your excellent video with unbiased comments. I’ve been with M3 for a year and half and use AP anywhere I can. These kind of video reminds me how amazing human brain with just a pair of eyes can do, instantly process recognition of all objects in sight even consider objects and situation from memory, prioritize them in current situation with 3D physics then instantly plots the best course and steer. Even our brain have a spare to enjoy music or engaging conversation!
Wow I have to say for only $1k and open source.. damm its much harder to support a ton of different automakers.. vs just one .. now I feel my Tesla should be even better as they are only working on one platform.. open source truly seems to perform better overall and with just one camera.. that is truly impressive.. Tesla should buy them out!
@@ericbell217 If basically one guy with sparse resources can produce OP which is about neck and neck with AP2.5 it really forces one to ask what Tesla's massive team have been titting around at with AP for the past 3.5 years since the split with MobilEye, because "progress" has been painfully slow, kinda like watching an arthritic frog traverse a lake of treacle.
Eric Bell watch the video.. if that system had equal amount of sensors, gps and cameras plus a faster gpu/cpu this open source would be well ahead.. as it is now with only one camera and at a cost of only $1k Tesla got nothing on it .. as I said Tesla should buy them out and integrate this platform with all their sensors, gps and cameras plus faster hardware they would be closer to level 5 by now.
@@fullyelectric that's false. Lane keep is a rather easy thing to do and that's all it can do. It follows so far behind cars on the highway that's its not even practical in many use cases. Even then, it swerves back and forth on the road whenever there were straight cracks on the road. It's good for 1k, but it's laughable you think it's as good as auto pilot in any way and definitely not worth acquiring. I'm not sure what benefit they'd gain over the current system.
Openpilot is great. I ended up buying a 2020 corolla for this purpose and I’m not the kind of person who likes buying new cars. I regret not getting it sooner Openpilot is priceless and it should be worth more.
WOW! One of the best objective evaluations of auto pilot I have ever seen. I have two Tesla s with FSD, but openpilot is very impressive for how it well it does for the cost and simplicity of hardware requirements. I was never going to buy a an internal combustion car ever again, but Openpilot is an effort that I want to support. am planning to buy a plug in hybrid Toyota RAV 4 just to support the amazing work that Comma AI is doing.
You don’t have to have your hand on it the whole time. You can take your hands off and when the blue signal to touch it, you can just touch the volume button up or down and it goes away. So it really just wants to make sure you’re paying attention.
Andy, At 9:20 when trying to initiate a lane change and you had to do it manually, it may have been due to an illegal lane change on a solid white line being prevented. I know everyone does it, but police will issue a ticket if crossing over a solid white line.
This was a great head-to-head real world test. I learned so much about the capabilities of both platforms as at 13 Jan 2020. Really enlightening. So grateful!
@@howerover in Tesla they claim their Autopilot have 2 system running in parallel for redundancy, this is what OpenPilot should also try in my opinion..
@@davidcazares7441 because that's literally what they announced when they released the requirements to get OpenPilot up and running. They've also added support for the LeEco Le Pro3.
Do you know why Openpilot decided to only go with one camera? This seems to cause most of the problems with it, and if its a cost thing, I'd be sure many users would pay an extra couple hundred dollars as a bonus upgrade for safety.
Mainly to focus on developing the software stack and trying to do as much as possible in building a good machine learning model for driving. In the future they may expand to more cameras but for now it's a 13 employee startup working very hard to improve what they currently have software wise! So much left to do with just single camera systems!
18:14 -- as far as I know when using OpenPilot if you press the accelerator then it will *disable* the 'self-driving' system. In Tesla Autopilot 'self-driving' system you can press the accelerator and the system stays active.
Openpilot disengages on acceleration or braking; it does not disengage if the driver turns the wheel. Autopilot disengages on steering or braking but not acceleration.
@@aslye , I was pointing out the difference with accel. Certainly, both (and all) systems would disengage with braking just like basic cruise control has done for decades. Very good point about Openpilot not disengaging when the driver turns the wheel while Tesla Autopilot would disengage AP(Autosteer) but leave AP(TACC) engaged.
@@DerekSmit Except they can't run their car on a road they haven't 100% mapped out. The way they are operating is completely unrealistic, and I don't know when they will stop working with fully mapped out roads and move to a dynamic model.
@@kendokaaa Nah, if you can do it with cameras, which is more difficult to achieve, you don't need lidar. Lidar is a crutch which has a ceiling, doing it entirely with cameras is the ultimate in cost effectiveness.
@Clutch Solution so you're claiming that cameras can see better in low visibility conditions than lidar, interesting. I didn't know cameras could see at a good distance in darkness and/or rain
Regardless of how well they both work, if you have to keep your eyes on the road, then you might as well drive yourself. Personally, I'd much rather have a HUD that flags potential threats in a non-intrusive way. This can be used to collect massive amounts of study cases for machine learning too.
I agree. I test drove a Tesla and regardless of how good it was at the time, it felt a lot like letting a learner driver drive. It should work well but you still have to keep your attention in case it doesn't and that I find more stressful than driving my self. I find it takes more effort from my part because I have to look further ahead than I normally would then watch what it is doing and then try to predict if it will do the right thing. When I drive all thus happens automatically. I do not need to think or double guess what the system us going to do. In my opinion when I do not need to pay attention and it will feel like a competent driver I will not use it. For highway cruising or slow traffic I think it will be ok.
Honestly it looks like if OpenPilot starts using multiple cameras it'll easily surpass Tesla autopilot in every scenario. I'm surprised that I've never seen much hype around such a great utility.
@@outcast566 so will openpilot speed up if it spots a car zooming at it to the rear? th-cam.com/video/h3Dzi0_PmmU/w-d-xo.html plus 10 million miles is a 1/100 of tesla, not that impressive compared to over companies like google waymo
@@TypicalBlox AS it is now Tesla is better. But we can't overlook the fact that Tesla cars are factory made with autopilot compatibility. They use multiple cameras and radar. OpenPilot is using just one camera. Give it access to that many hardware and I don't think it'll be hard for Openpilot to match Tesla autopilot.
wow, what a great video, Watched all 30 mins of it. At least now we can see what owning a autopilot car is like. Maybe someday we'll be able to buy one. Thanks guys for the upload.
From what I hear it works great. These new cars have these super bright LED headlights that look like fucking high beams blinding oncoming traffic in the night lol. So OpenPilot should see just fine.
I haven't been able to drive for years because of vision issues. Nice to see the current state of software-driven cars, especially the open-source solution. When I can get a car to drive me from point A to point B that will finally be something. Until then, I really hope some bad actor doesn't find a way to send their own OTA updates to people's cars.
Tesla following distance is not “car lengths”, btw; I believe it is also time-based (e.g., sensitive to speed). At setting 7 (max) it is a bit more than 2 seconds, which is the recommended distance.
WATCH UPDATED VIDEO COMPARISON ➡️ th-cam.com/video/Hc8G5U0p1sA/w-d-xo.html
🚘 TESLA RESOURCES
Best Model 3 accessories: geni.us/tm3accs
Rent a Tesla: geni.us/Turo
Buying a Tesla? Free Supercharging: geni.us/t3sla
Geeky Tesla shirts & prints: geni.us/SFSF
My best Model 3 videos: geni.us/cjUF9g
⭐️ FAVORITE RESOURCES
My TH-cam camera gear: geni.us/YTgear
TH-cam Success course: geni.us/YTcourse
Real estate investing: geni.us/Fundrise
Stock investing app: geni.us/qZkq
Best savings account: geni.us/Wealthfront
My Amazon store: geni.us/AMZN
👇 JOIN ME
TH-cam: geni.us/AndySlyeYT
Twitter: twitter.com/slye
Instagram: instagram.com/slye
Email squad: geni.us/TeslaFan
Andy Slye great idea to put this information as a post vs just in the description that I never really take the time to go open and look at .. it’s a PITA on some device..
Andy - Open pilot lacks the Basic Safety systems found in any Tesla with AP3.
also, 8 cameras and RADAR and SONAR trumps just one camera.
Open Pilot cannot use ANY of the Built in Cameras in your car.
TESLA can Automatically detect cars in FRONT, Front Left, Front Right, Both sides, Both Bind Spots, Rear and Rear Left , and rear right.
Open Pilot cannot warn you that you are About to be Rear Ended.
the TESLA RADAR can Actually see the Road Ahead even Underneath cars in FRONT of you looking for Road Hazards.
Tesla AP3 is Level 3 - 4 , its why its called Autopilot 3.
Tesla AP3 will be Improved over time and TESLA is Already working on AP4 chips.
and Open pilot does NOT have Billions of miles Driven.
TESLA AP3 FSD is the Worlds most Advanced Autopilot.
Open pilot cannot slow down the car to a stop if you have a Heart Attack or other medical situation.
Lets see how well Open Pilot does on a 5,000 mile trip.
that test is EASY for a Tesla.
Lets see how well Open Pilot does in DUST, FOG, or SMOKE ?
RADAR negates the lack of vision, allowing you to FLEE in a Natural Disaster.
i would never trust toyota adaptive cruise control, it might take you into a river.
@@markplott4820 I am a TM3 driver but I have to correct you.
First there are fours versions of Tesla autopilot: HW 1, 2, 2.5 and 3. It is not called HW 3 because it is level 3 and indeed it is not. It is currently level 2 and it will be level 4. HW 1 is using the MobileEye computer and only has one 1 camera. HW 2 is Nvidia with one GPU and 8 cameras. HW 2.5 adds an extra Nvidia GPU for redundancy. HW 3 swaps out the Nvidia GPU for a Tesla made neural network processor. HW 2 and 2.5 can be upgraded to HW 3. HW 1 can not be upgraded.
Three cameras all look straight ahead. The difference is the angle of field of view (one camera looks far ahead in narrow field of view, another has a very wide field of view and the last somewhere in the middle). Two cameras on each side of the car, with one camera looking forward and to the side, the other rear and to the side. One camera at the back of the car which is NOT used by autopilot. That one is currently only to help park the car.
Autopilot does not currently use all these camera to their full extend. It is mostly using one forward facing camera, which is why it also sometimes has trouble with very tight curves. It simply can not see the road even though it should be able to, if it would just switch to a side mounted camera or the very wide field of view front camera. It does use the side mounted cameras to detect traffic during lane change however.
Think about it like this. For each camera Tesla trained one or more neural networks. The one running on the front camera is tracking lines used to center the car in the lanes. The neural network running the side cameras however were not trained for lines. Instead they trained that to track cars and objects, such as traffic cones.
And by the way, autopilot disengages during bad weather. You can not even use the radar to get old fashioned cruise control. Technically that would be possible but Tesla does not allow it.
@@BaldurNorddahl - All current Model S/X/3 are AP3 computers since April 2019. HW 2.5 since September 2018 can get Updated hardware in the AP3 chips.
Both HW1 and HW2 are Legacy systems and cannot get Hardware Upgrades but the can still get Software updates.
AP3 is Level 3 - 4 Period, it can Navigate on streets, to the on Ramp, Speed up + change lanes, pass SLOW moving cars, Navigate to Exit, take off Ramp and drop you off at work, this is possible with AP3 v10.
this latest Software update is much improved since AP3 first came out, v10 is a game changer.
a Friend of mine Moshe the Electric Israeli recently completed a 3,000 mile Road Trip completely using Autopilot.
and I am sure AP3 v10 can also do a 5,000 mile Trip.
Police: Do you know why I pulled you over
No Sir I was Sleeping
How dare you assuming the gender!
@John Buick Good stuff!
Just realized self driving will put a stop to a lot of ticket revenue gathering. What will they do to get their money after that? The same thing? Will they get fired? Hmm.
@John Buick It's referring to the big argument going on right now around with Jordan Peterson... It's just a joke if you follow the news on the gender names battle. It's been going on for min 2 years.
What seems to be the officer problem?
i miss youtube videos like this just 2 normal guys talking about something they are passionate about
Thanks for watching! We may do a future comparison video so stay tuned and subscribe here: geni.us/AndySlyeYT
I miss leaving TH-cam comments, just me leaving a comment
...and letting their cars handle the driving. Good old times. :D
It’s fine, just go to Logan Paul Vlogs for a video on two guys about their passion towards watching Lana Rhoades. TH-cam 2020 is on a great start
You’re homophobic
The Openpilot project is amazing, but I think it needs a rear camera and additional sensors to be really suitable.
This is probably the reason that George Hotz had his peepee smacked by the NHTSA a few years ago. I've noticed that this software seems to be limited to cars that already have sensors installed. For this to be a truly revolutionary thing, it would have to work with aftermarket sensors and include a suite of calibration tools ... and possibly even require professional installation. Simply strapping a cell phone to your ECM and pointing your 2-ton missile down the highway while declaring "IMMA ROBOT NAO" is not sufficient safety testing.
@@chooseymomschoose "it would have to work with aftermarket sensors and include a suite of calibration tools" it is open source, whats stopping you or anyone else from coding that feature! I predict that the rise in demand for self-driving cars, and the fact overtime, features tend to cost less, it'll be more common to see these new cars ship with enhanced driving capabilities, allowing a larger group of people to take advantage of commai
OK that's the downside of this as far as I can tell atm. My car allready has a good ammount of features and sensores that make is almost capable of self driving. I have lane assist, blis, forward assist, cruise control, camaras in the front and back and all of that stuff. Basically all it would need is a software that brings the informations and features together so it would drive fully autonomous. I mean, with cruise control and lane assist enabled it allready feels like autopilot the problem is that lane assist is designed to only work and take control if you're going to cross a line instead of "fixing" the car in between the lanes ...
and wider lens maybe
@1234 Exactly. From my perspective, you don't even need extra censors, add rear view camera(s) and you're good. Humans already are only driving with their eyes, as long as the system can see the cars, it would still react much faster than we can and provide way better results. All these fancy sensors in Tesla's seem to be a more reassurance and increase price than necessity.
How did I not know Openpilot existed?? I just went down a rabbit hole learning as much about it as I could! Thanks for the awesome video!
Glad you enjoyed it! And yes, OP is an amazing technology for non-Tesla cars.
Geohot is awesome, check out his appearance on Lex Fridmans podcast
@@Phoenixklinge Geohot created this?? Man that brings back the PurpleRa1n days GeoHot was the jailbreak king
lol same i have a 2018 corolla and i think i know what I want for my birthday lol
The same way that you don't know about wayve, the system being used in the Jaguar i Pace.
It uses machine learning.
The original was developed using the Renault Twizy.
th-cam.com/channels/NERNUuB5kAj7rGGfhk0C3A.html
Very awesome comparison. I am impressed with OpenPilot--just wow!
Thanks for watching! Openpilot sure is incredible for being open source and $1k
great product if you dont live in a congested city .
@@aslye very impressive to see it and thanks for showing what is possible,
BUT there are still big problems and not at all impressive how you drive or let the car drive and run "dark-orange" or red lights (which we jokingly call "cherry green" :-)
about the "open system": there is already quite some discussion about the responsibility on "official" systems, and they are equipped a lot better with many more cameras, sensors, radar, lidar, etc. having such systems on many "less suited" cars will increase risks, and might make it more difficult to increase acceptance of any(!) kind of self driving in the public opinion and in changing laws to make them legal. afaik, at least in europe, such "private modifications" currently are not legal and invalidate any permission to use those cars in public traffic, as well as invalidate the mandatory insurance.
in addition, if it is really "open source" and everybody can modify it and make his own rules and determine the degree of aggressiveness, it will be turning even worse and not make selfdriving cars safer in general.
about the traffic lights: on the widening lanes, you are so concentrated on watching how the car handles the widening lanes that you forget to check the traffic lights. several times the light switches to yellow at least three seconds before you reach the crossing, eg at 6:42 the lights turn yellow and when you enter the crossing at 6:45, the lights just turned red !
i don't live in the usa and maybe you still may cross roads like that, but where i live you have to stop in front of the crossing when you safely (depending on speed and traffic behind you) can break. three seconds should be "good enough" to do that, and on your red, other cars might already try doing a jumpstart from the sides before you finished crossing 4-5 seconds after first seeing "yellow" ...
@@Anson_AKB those systems rely on also not braking unaturally fast, because then the programer would get blamed for getting rear ended. just as humans, AI can get blinded or miscalculate LOL
@@Anson_AKB I encourage you to watch some talks from George hotz. You make a lot of false assumptions in this comment and I think it would benefit you to learn more about adas systems and what is possible with current technology.
Excellent side by side guys. I am totally impressed with OPEN PILOT, but, as a rule, I never use cruise ctrl on local streets like those you traveled on anyway. So, I am totally blown away that both system even worked at all. This was soooo Eye-opening and informative straight up. Phenomenal work guys, and please do more analysis like these in the future.
You can tell the Tesla was trained in the Bay Area since to doesn't let people into lanes lol
lol
+Steven Smith Here in the UK the roads are really busy so often people don't maintain the distance they should... It's quite common to see a Tesla with its blinker on to change lanes, then time out, then the blinker is back on, then it times out again. At peak times the Tesla will never find a gap big enough, whereas a human driver might take a greater degree of risk by pulling into the smaller gap.
@@donkmeister as an American driver I always found European roads way too cramped.
lol
They use all cars to train the neural networks, no matter where they are. But they definitely have a lot of data from the Bay Area.
Really impressive it's able to keep up for the most part with just a single camera.
TESLA has 8 plus Radar
@@markplott4820 Also 12 sonar sensors.
Most ADAS systems use a single camera + radar. You could get away with a mono camera, with less redundancy, but then again, comma.ai isn't exactly going to get in hot water with the NHTSA.
Now imagine if that cheap piece of tech adds additional data inputs. You can probably sleep on it and wake up parked in your destination.
@@johnhoo6707 redundancy is a good thing when people's lives are at risk.
My two reactions: (1) Paul Rudd is bloody everywhere these days! and (2) the one thing Openpilot has over Tesla is you can take it with you to the next vehicle.
As long as the next vehicle is supported by Openpilot: comma.ai/vehicles
Not exactly. I have a 2015 Chrysler Town ad Country. Not compatible.
Not for long, since in 20-25 years from now every vehicules will already have autopilot in them.
Cool for now, but not in long term.
@@mackblack5153 Autopilot is a gimmick, never going to be perfect. AI is not perfect because we are not perfect.
TESLA USES REAL WORLD INTERACTIONS, WHICH ARE FAR SUPERIOR THAN THE COMPUTER SIMULATIONS WHERE INPUT MUST BE ADDED. THERE ARE WAY TOO MANY WIERD THINGS THAT GO ON LIKE A FULL WHEEL TRAVELING LIKE 30 MPH OR SO HEADING STRAIGHT AT YOU ON A HIGHWAY.
I feel like I would be much more comfortable having the view the Tesla has because it shows that the computer is detecting everything around you where the other system you just have to trust that it sees things but there is really no way to verify.
It's hard to tell in this footage but openpilot shows the detected lane lines and path as well as the lead car being detected on the screen mounted on the windshield.
one word ... Summon
Teslas view... well...it doesn't show what the computer detects. It shows what the computer THINKS it has detected, and nothing more. And that is a problem, especially since YOU as a driver should have your eyes on the road at all times, and NOT be looking down on a screen. It makes it Less safe to have a lot of distractions in the car.
@@Xanthopteryx But openpilot is literally a phone in the way of the windscreen...
@@ItsaB3AR Yes it is but:
It's placed so that you do not have to move your eye so much and, it doesn't show a lot of fancy icons (cars, trash bin(!?), cones, ...). And, OpenPilot watches you, so if you look away too long, it will nag you about it.
Love openpilot on my Toyota Prius Prime. Poor man's Tesla :P
that's interesting. I'm in Japan and I have a 2015 Prius and mine came with those features but didn't need any kit.
@@ryanDogAdventure Toyota did have some good features in the gen3 with advanced tech package. The system in the gen4 and Prime got worse and is basically useless. Openpilot is a much needed upgrade.
@@techie163 that must be outside of Japan. As of last year, almost all brands, Subaru, Mitsubishi and such have level 2 systems in them. I think Japan is gearing up to making it mandatory for cars due to seniors causing accidents. They are really focusing a lot on this here in Japan.
Nathan - you are Better off with a TESLA model 3, they can be Leased and be bought for $35,400 brand new.
@@markplott4820 True, that wasn't available when I bought my Prime in 2018. Maybe my next car!
Awesome comparison. If they could just add a few more cameras this would be killer!
Hi I watch your videos
Yes really impressive for a single camera Android phone
Yes. But... Then they would need a far faster processor set to compile the different views.
@@loungelizard836 That's not the way neural networks work.
@@pimpstick2 LOL. That's not the way neural networks work man... You think open pilots hardware (which is basically a phone) is running a neural network? No, its not. Lounge lizard is correct that you would need faster processor when using open pilot on your car if there were extra camera angles. The neural network is what compiles data (video) from every car, mashes them together and compiles a driving model or software update that then goes out and is sent out to all the phones to run. Consumer hardware aren't running the neural networks.
Currently open pilot will never be able to evolve past level 2 autonomous driving unless in the future they are able to add extra hardware and cameras to the sides and rear of the car. They currently have no plans to do this yet and are mostly just focusing on perfecting L2.
As a Tesla owner, I think OpenPilot is great. Not that it’s more capable, but that the more cars that have some kind of drivers assist capability, the better. Unfortunately, the one ICE vehicle we still have, a Chrysler Town & Country minivan is not on the compatible list yet. If it does become compatible, I would add OpenPilot in a heartbeat.
Unless your Chrysler minivan has a motorized steering rack, it unfortunately will never be compatible. OpenPilot can only support vehicles whose steering rack is motorized.
George Hotz doing great work with Comma AI - dude has a cool story and a great interview recently with Lex Fridman on Lex’s channel
Yep he's doing great work with OP
Musk should have hired him when he had the chance. Of course there would have been
personality conflicts that would have ended the relationship pretty fast. Hotz would have
advanced Tesla's vision system in months, rather than the years it is taking.
ram 1901 I definitely think he could have been as asset, but I agree I could see their personalities clashing lol. And I’m not sure he would revolutionize what Tesla is doing as he admitted himself that Tesla will win FSD in that interview. Far more visualization tech is ready at Tesla but they need approval before each new release by regulators
never knew there is something called the open pilot until this day. Great job guys?
Whomever designed some of these roads should be publicly shamed, jailed for a year then forced to correct them with a number 2 shovel.
Indian roads: hold my potholes
Those roads are in very good shape. Lack of lanes and patches for humans is not a problem if the autopilot is incapable of driving in such conditions than it is definitely not good enough to be adopted.
*whoever
@@yellnitroy2031 That's assuming there's enough contiguous road surface to have differentiable potholes.
Hi, Indian here.
So awesome that both of you fairly and honestly assess the two products. It would be great to have some competition between the two so that both can up their games.
We just bought a Tesla Model 3 and we love it! We are taking "baby steps" in learning to trust the vehicle. Pretty trippy!!
Would love to see OpenPIlot lead the way more as having a lead car seem very important to openpilot as compared to autopilot.
We can definitely test more without lead cars during our next comparison test. Thanks for watching!
This was absolutely fascinating! Thank you guys for taking the time to put Tesla Autopilot and Openpilot through their paces. I had no idea there was an self-driving aftermarket option like Openpilot; I'm impressed with how well it performed! It will definitely factor into my next purchase when it comes time to get a newer car. Please do more of these! Would love to see how they compare in situations like heavy traffic, busy parking lots, even dirt roads!
Thanks for watching! We'll for sure try to do another comparison video later this year or early next year.
Amazing comparison! I've been using openpilot for years, even before they sold any products. It's been amazing to watch my Honda Civic get better and better with ever openpilot update. I almost never manually drive during road trips now!
What year model is your civic?
@@jaunie8217 2017 EX
@@VirtuallyChris nice!
Tesla is also timebased distance not carlenghts.
Yes that’s correct. Sorry for confusion
I think that is also the same with the Corolla
Openpilot is open source, it is more like android vs apple
Yep in fact Openpilot founder said something similar. If Tesla AP is the Apple of autonomy he wanted OP to be the Android of autonomy
Great, so it’ll be a project that either forks or gets abandoned? Jk, I wish it all the best.
Andy Slye I scrolled just to make sure someone had commented on this
@@JPatel1995 android is better than apple now
@@JPatel1995 🤪 keep being a sheep
I think AutoPilot is functionally better, but OpenPilot is more technically impressive due to the insane simplicity of the hardware yet it still copies a lot of AutoPilot's functionality.
Yes
I was also surprised it did this well.
I'm just speculating, but I wonder whether the limited data (just a single camera) makes it actually easier to train those NNs for the basic cases, but over time it will start losing behind Tesla, which will be able to evaluate more complex situations better.
@@peter.g6 You're probably right about that. Right now, it's easier to catch up, but it will eventually lose out. Even though Tesla's system is more complex and more difficult to train, it's also, well, precisely that - more complex. That means it will be able to better deal with more complex situations once it takes the time to be trained.
I'm not sure it sopies Tesla. Openpilot has been around since 2016 now (and earlier as prototype by the developer), and they take it their way. It's not about copying something - it's about getting it to work.
Copies? Bruh they have the same goal. Similar function doesnt mean copying.
What a great video. I use OpenPilot on my Volt and it's mind blowing how good it is for being a fairly low cost cell phone in a 3D printed case. Comma has done some incredible work.
For the follow distance issue: Since a a large amount of OpenPilot is open source there are a lot of community forks that add features. I use Kegman's fork which has 4 distance settings. Makes a huge difference. I can also disable the steering and just keep speed adjustment enabled. This is really useful for construction zones or weird roads with bad markings. I can take over steering and let OpenPilot handle speed. This all integrates nicely into the car. It uses the car's built in LKAS button to control whether steering is enabled or not.
I am jealous of his Toyota's steering torque though! My Volt needs a bit more help on sharp curves especially at lower speeds.
It’s not available on every car yet I’m waiting so I can install
@@Flankerr The cell phone component was $500, not $1,000. I could have built it for cheaper, but I didn't feel like messing with that. The total cost of the system, including custom GM stuff, was $1,000. I also could have built the GM wiring for cheaper, but again, it wasn't really worth my time. I would have gladly paid Comma $1,000 for just the cell phone portion of it. I get enough utility from it and it's worth supporting a small, innovative company. The driving model is not open source which is what most of their research goes in to.
It should also be noted that this isn't a real product yet. They're selling dev kits to help fund the cost of development and get community support. These should not be purchased by regular people yet. They require a fair bit of tinkering to get working well. They'll probably come out with a product eventually that's much easier to get running.
@@Flankerr: Damn, Tesla Autopilot is many times that expensive and in the same class, re function. After ALL this time, false promised by Musk, hardware upgrades, etc.
Great video. I've been curious about a comparison of the two system. George Hotz the Founder of Open Pilot seems very dedicated and passionate to bringing this system to as many drivers as possible. Once he and his team succeed in perfecting the system. I'm sure they will license it to traditional car manufacturers.
George Hotz is a genious. I watched alot of his livestreams and he said whenever Tesla solves level 5 autonomy comma ai will solve it too after 2-3 years. No doubt though that Tesla will be the first of all the companies trying to get to level 5.
Not sure if this is still the business plan but GeoHot recently said this on Discord:
"We’ll leave the hw game in 3-5 years when Xiaomi and Huawei are making Openpilot devices (based on Qualcomm or NVIDIA chipsets)
There will be plenty of 3rd parties making hardware by then, some with our blessing and support, and they'll be able to use trademarks like "openpilot" and our logo ... such devices will be allowed on our network and will be supported with upstream OP also.
OP would be just a software update ... we want to control the OTA on cars and use the OEM’s compute and have a monthly subscription business model.
There will also be devices without out blessing, branded differently, not on our network and requiring you to run forks, exactly the same as Android.
Still gonna be 5-10 years before OEMs get the picture and include it 1st party."
Source: i.imgur.com/GKdv3IH.png
@@foobarrel9046 I thought he left the company.
Yes I want to see more videos like this and I want them to be 100% honest and no fake it till you make it. Good job well done thank you
Honestly i like how careful openpilot is especially during heavy traffic. I cant wait for self driving to become a thing.. I cant drive for medical reasons (NEVER have been able to and im 32 now) so my life has been SUPER limited. can't get jobs i want because i have to work within 5-10 miles of work or else im biking longer than im working. I cant go get groceries easily (though now people deliver so thats cool). i cant even hang out with friends really without paying money for a taxi or relying on a friend.
Wow hold on here... Open pilot never plans to be automonous and is meant as a simple lane scanner to follow said lanes. It has one to none safety features while autopilot has a group of cameras giving you 360* coverage and full automonous driving in the future allowing people who can't drive have the computer drive for you
@@uhsund huh? The point of the comment isn't to compare the two.
@@UchihaSumairu but the video you were commenting is. And good luck with this trash in thr future. tesla.takes the win here.
If you can bike, look up for velomobile. People have reported to be able to maintain 30mph with a reasonable physical effort (
@@chuck_norris that's cool if it does. I don't have either. It isn't a competition. Are you b8?
@16:45. Actually having that much of a gap between cars is ideal given your speed. It's this mentality of not leaving long enough gaps that cause traffic and accidents.
Yes. The driver is tailgating. Tailgating is dangerous, especially on highways: the braking distance (and kinetic energy of a potential crash) both increase with the SQUARE of the speed.
@@DanHaiduc There's so many darn tail- gators in the US!
wow great comparison! it makes me very happy to know that there is a open source option out there.
Thanks for watching!
Great comparison. I’m so geeked you drove literally to my workplace. I take the same route to work every morning and I was so confused on why everything looked so familiar to work, until I realized you really drove through where I work 😂
Nice!
Thanks for doing this comparison video, Andy. I heard George Hotz speak at our company's conference in December 2018 and met him personally. Comma AI's stuff is really great and I hope more car companies will start partnering with them. Just couple of quick things I wanted to point out: 1) Tesla's following distance isn't car lengths but it's "time-based" (not seconds, either). 2) Tesla doesn't recommend anyone using AP on city streets or windy roads (they state this in the manual), that will be available when FSD is developed and Tesla is rolling it out. Excellent video as always!
Model 3 specifically states in the manual that it isn't measuring in terms of cars but rather seconds. When you set it to "2" it isn't two car lengths, but 2 seconds of space. Check out the manual, someone showed it to me when I made the same mistake!
As a Tesla owner, I think Open Pilot looks excellent. We need as many cars as possible to be on systems like this to reduce accident rates. Having an affordable system that can be added to any car seems a great step in the right direction.
Except they cause more accidents then reduce them….
Open Pilot is quite amazing considering it is running on one camera and a phone. Well done!
yes it is almost haf as good as autopilot 👍🏻😘
@@chuck_norris Elon is globalist - I can't support that. I really want OpenPilot to succeed.
Man I've been dying to install OpenPilot in my Civic and this video made me want it even more. OpenPilot has improved a lot. I'm impressed.
Thanks for watching!
Civics have waay less steering torque compared to this video but it has enough for highway driving. Stay tuned though because civic firmware hacks that increase torque and bring steering to zero are going to be released soon if you're adventurous enough.
@@edwin260 oh I already know the steering torque on the Civic is garbage lol. I've seen a few people pull off firmware hacks for that motor though, is there any dev that planned on releasing to the public?
@@1blackbirrrd Check out the official comma ai discord. They recently added a firmware modding channel if you want check on the progress. The devs (not comma employees) briefly released a guide on how to do it with the necessary scripts on github, but they decided to take it down in order to refine it. But public release definitely coming soon.
@@edwin260 Awesome, thank you!
The only thing I didn't like about OpenPilot(owned for 2 weeks prior to buying Tesla) was the fact that you couldn't leave the ION Phone/Unit in the car due to heat issues. And your compatible car would only be as good as it's assistive cruise and lane keep assist capabilities.
Yes it has some limitations like that
You can now leave OpenPilot in your car 24/7 in Phoenix summer heat with the Comma Two.
This is a really big deal! Let's look at how this terrain could work with Tesla:
1) We will need entry-level products as a migration points. Many people are up for spending $1k to test out a technology. If Tesla's is vastly superior then after confirming "the concept is good but I'd like better" something like open-pilot becomes a bridge for Tesla buyers to want and value superior autonomy.
2) If Tesla will partner/produce a $1.5k version like this that is as good or better than open pilot - or a partnership, there is a huge win/win: a) Safety for drivers should improve, saving lives, building Tesla's brand. b) Tesla can gather more miles driven data and learn how to make it's main model more efficient and measure the safety discrepancy per million miles, leading to many contributions to perceived values. c) By far, Telsa's largest asset is owning and selling/updating the safest most versatile autonomy software in the world. Most of the value will come from subscription services to as many cars as possible on the road. Tesla can sell design-config and software support to private drivers and legacy auto-makers as a way of keeping customers. This is pure profit on Tesla's bottom line.
3) If Elon does not embrace non-Tesla partnerships and open-source, a partnership of the main legacy automakers, venture investors and home-kits like this probably offer the biggest challenge to Tesla over time. The autonomy piece is huge, and much like android took over apple in market-share because it collaborated rather than controlled, the collaborative open source block-chain model of sharing data is the ultimate winner in all spaces.
4) I'm excited to try and fit my tundra with one of these set ups, which I'm sure will become more and more versatile over time. Thanks for telling me that they exist! I'm happy to spend more for better safety so I really hope that this becomes a huge industry: a) People making videos on the safest, tested set-up for every vehicle ever made (For the Tundra 2001 you want five cameras from this source mounted exactly here in this way, then you want the screen attached here, and a subscription with X company and a record set-up with X hard-drive.). Once this get's proven (Tesla currently has 8x safety factor over humans which is enough to get it but not enough to get a new car if you don't have a Tesla and like your cars features) then literally everyone will retrofit existing cars and all new cars will have these features. The exponential tech-price curve dictates that every few years something twice as good will be available for the same price and the profit margins are so high that competitors will enter the market until you can buy safety per mile for almost nothing and we stabilize at 7-12 times safer than good drivers globally. Tesla has the lead, but eventually all companies will crack this one way or another very cheaply because of the exponential learning of A.I., hacking, intellectual property theft, better simulations and time. It will be a commodity with few distinctions in twenty years for a few cents a mile to have top-line A.I. guiding all transportation including planes, boats etc. because it will save money and be safer.
Andy, I appreciate you showcasing and giving points to a cheaper competitive model given your passion for Tesla. That makes you trustworthy as an information source.
Sincerely,
Dane
"Off-ramp slow downs" are actually based on the speed that other Teslas drove around that same corner, so autopilot references what other Teslas have done and uses that to set the speed on that same corner. From page 87 in the Model S manual: "Off-ramp speed: When enabled while on a highway interchange
or off-ramp, Traffic-Aware Cruise Control may
reduce your set speed ...to as slow as 25 mph (40 km/h) to better match the reported speeds of
other Tesla vehicles that have driven at that
specific location."
Openpilot seems really neat. Reminds me a lot of Autopilot V1 in my 2015 MS. Reacted to tight turns and such similarly
Great collab! Very insightful for folks that own Openpilot or Autopilot to watch the differences.
Thanks Michael!
Could you make a new comparison with the newer versions of each one? (comma three)
+1
Yes, a follow up would be nice.
(comma three)
This is one of the best side-by-side comparisons of autopilot systems I have seen, with two drivers and two vehicles following the same route. There's a very good step-by-step replay and commentary about how each system handled various challenges. I am really surprised how a single-camera, smartphone-based system can compete so well with a much more layered technology in the Tesla.
Great comparison between AP and OP! My wife did a presentation at PepsiCo a while ago about AI Machine Learning and she did mention both Comma.ai and Tesla. It was my first time hearing about Comma for quite a while and am happy to see how well they've been progressing!
That's great to hear, Julio!
I drive AP and have paid for FSD. But so far I vote for OP. OP wins at the moment especially if cost is considered.
Wow I had no idea openpilot existed. Will be getting this with for my next vehicle!
👍
thanks for proofing how good Tesla's autoPilot has become!
Thanks for the vid. I have Garmin GPS that tries to warn of lane departure and it gets confused with tar strips as well. It also does collision detection and notifies when the car in front moves (for stoplight texters). Single camera, it is pretty impressive. It does speed notifications and traffic. You can also add another camera via bluetooth. It would be cool to see the GPS technology married with the opensource stuff.
You two make a great team and we need all the help we can get in autonomous driving, keep these videos coming THANK YOU both!
I used Toyota’s autopilot version on a new Corolla Hatchback and was impressed how good it was driving by itself on the highway
Right?
I didn´t even know Openpilot exists. I am positively impressed by this device. It deserves all respect for its functions and affordable price. Thanks for the great video!
Thanks for watching!
Comma 2 just came out! I just received it!! Thanks Andy for this video. I was definitely convinced to get it for my 2018 camry
Good video. Surprised to see openpilot performing just as good or better than Tesla in some situations. I considered it when I had the Chevy Volt, but I'm not really a DIY type of person when it comes to my car.
Yes it was my first time seeing Openpilot in action and it was quite impressive. It's a no-brainer for people with a compatible vehicle model IMO
Really well done video Andy. I had no idea there was an open source openpilot available for other cars.
Btw the blue steeling wheel icon means auto steeling. The blue circle around the speed means the advanced adaptive cruise. These two together is the basic autopilot now.
Really impressed by Openpilot
Thanks for watching!
@@aslye now put it on an electric car. unless it already is and i didnt hear it.
OpenPilot is amazing. I mean autopilot is the superior hardware/software here but then again it has loads of cameras and native support from the ground up. OpenPilot is a fraction of the cost and basically plug and play for compatible cars. Amazing tech with very, very decent performance.
Nice to see both systems side by side. Openpilot is impressive. Wonder how Mobile Eye does compare those days. The more competition the better. Thanks for your videos.
Thanks for watching, Frank!
MobilEye is far beyond Tesla's Apr. 2019 FSD demo, but they did start ~10 years earlier; see this recent 25min drive through Jerusalem: th-cam.com/video/hCWL0XF_f8Y/w-d-xo.html
Nice comparisson video, im also driving with open pilot now. Really makes driving less exhausting.
Wait. Corollas can drive themselves? The world is amazing now! Dogs can talk, people can fly, & cars can think. When did all this happen?
SalvadorMonella yesterday :-)
It's not the Corolla.... It's the open AI pilot. That's kinda what the _whole video_ is about.
Sergio Martinez I have the 2020 Corolla Hatchback and it has lane assist built in. Mostly just works for interstate travel but it’s pretty good.
@@thisisntsergio1352 Yeah, but the Corolla has the capability of using OpenAI? That means the Corolla has servo-motors connected to the steering, the braking, & the gas! Corolla is one of Toyota's lower end cars. I'd say down from that is a Yaris, so Corolla would be in the bottom two. When did *that* happen? I guess about the same time dogs learned to talk & people learned to fly unassisted. I drove a Camry that had lane assist, and that was awful. Turning on lane assist was more akin to bowling where they put those long pads in the gutters so kids can't throw gutter balls: I ping-ponged back and forth in the lane. Automatic driving isn't like that. I like the idea of OpenAI. I run Linux & use opensource software all the time. But having that capability means all the other pieces of the self-driving puzzle have been put in place, which astounds me. I used to have a Corolla. I drive Tesla (& others), and find it hard to imagine a Corolla can compete. Plus, Corollas aren't electric, and as much as I wish it weren't, being all electric is a huge deal.
@@phillipcrittendon9172 I drove a Camry that had lane-assist and didn't like it at all. I know a lot about Toyota. Way more than 99.9% of the people out there (I provided critical consulting to their US operations for years) and I think they're in trouble.
The fact that most have no idea that other manufcaturers have similar competitive systems shows the value of Tesla's PR hype machine . It is, if not accurate or truthful - at least supremely effective.
Superb video. Great dialogue. Well done guys.
Thanks James!
@Andy Slye , you keep saying Tesla Autopilot (TACC) lets you set car lengths as the following distance but they absolutely are not 'car lengths'. The AP (TACC) setting of '4' is not the same distance at 70 mph as when you are traveling 35 mph. It is 'time-based'. Tesla manual: "Each setting corresponds to a time-based
distance that represents how long it takes for Model 3, from its current location, to reach the location of the rear bumper of the vehicle ahead of you."
TACC and autopilot are NOT the same thing.
@@MrUpscaleman But when it comes to following distance, they do the same thing.
@@MrUpscaleman Units are time not car length. Both TACC and Autosteer are a subset of AutoPilot. See this image: i.imgur.com/8D8IZRP.jpg
@@nZifnab That's NOT what you said...you're just trying to cover for a mistake now.
David Symons huh? This isn’t my comment thread lol
How can a informative video like this gets dislikes? If u dont like andy just dont follow him wtf, great job, I wish I could work in a software company that programs autopilots
That's awesome!!
Andy and Logan, you both should do 2.0 comparison video for Tesla FSD Beta and Openpilot 0.8. Both had their upgrades, so it would be worth seeing the improvements.
No offensive but that dude talking with Andy sounds like Toby from "The Office"
Amazing content by the way!
🤣🤣
That was my first thought too!
no offensive but why does every lemming have to refer to that and other zionist made stupid tv shows
@@abdou6003 wut?
god nooo
I've found that Teslas won't auto-lane-change across a solid white line, and has trouble with changing lanes with very tight dashed lines. In earlier updates, say maybe in November 2019, if you started an auto-lane-change when the lane lines were dashed but you started entering a solid white line section, such as in construction zones where they shift the 2-4 lanes to the left or right, the car would jump back to the other lane. Now, at least in 40.50.7 and a bit earlier, if the car has already started changing lanes, even if the line turns solid the car will complete the lane change.
I'm thrilled to see how far comma.ai OpenPilot has come! It looks like Autopilot wins in more conditions and offers a bit more of a smooth experience in most conditions, but it's great to see where OpenPilot beats Autopilot too.
Great video!
Honestly I went into this thinking Autopilot would beat Openpilot hands down, no ifs ands or buts, but using 1/8th the cameras, no radar or anything else this was very impressive! It will take a long time for them to get the miles Tesla has right now, but it's surprisingly effective for what it is!
It just proves the fact of how Tesla autopilot is overmarketed by Elon.
Man, this video kind of show just how ready, but also not ready autopilot is. Can't help but to be at awe how the human brain can navigate these courses without even really thinking about it. It's so amazing! Autopilot doesn't have to eat sleep or take a break. Its whole existence and design were meant to navigate these streets. and it can't even do it as we can. It's going to be uncanny when it finally can beat us by machine learning!
Thank you for your excellent video with unbiased comments. I’ve been with M3 for a year and half and use AP anywhere I can. These kind of video reminds me how amazing human brain with just a pair of eyes can do, instantly process recognition of all objects in sight even consider objects and situation from memory, prioritize them in current situation with 3D physics then instantly plots the best course and steer. Even our brain have a spare to enjoy music or engaging conversation!
Great video. Appreciate the honesty and non-bias break down of both systems. You two are fantastic.
Great video, Andy and Logan!
Great job Andy! I've done lots of Autopilot testing and its very cool to see Tesla's AP compared to another system.
Wow I have to say for only $1k and open source.. damm its much harder to support a ton of different automakers.. vs just one .. now I feel my Tesla should be even better as they are only working on one platform.. open source truly seems to perform better overall and with just one camera.. that is truly impressive.. Tesla should buy them out!
Tesla like Apple don't like opensource.
Better overall? I mean it's good for one camera but in no way was it better than Tesla in almost all the tests.
@@ericbell217 If basically one guy with sparse resources can produce OP which is about neck and neck with AP2.5 it really forces one to ask what Tesla's massive team have been titting around at with AP for the past 3.5 years since the split with MobilEye, because "progress" has been painfully slow, kinda like watching an arthritic frog traverse a lake of treacle.
Eric Bell watch the video.. if that system had equal amount of sensors, gps and cameras plus a faster gpu/cpu this open source would be well ahead.. as it is now with only one camera and at a cost of only $1k Tesla got nothing on it .. as I said Tesla should buy them out and integrate this platform with all their sensors, gps and cameras plus faster hardware they would be closer to level 5 by now.
@@fullyelectric that's false. Lane keep is a rather easy thing to do and that's all it can do. It follows so far behind cars on the highway that's its not even practical in many use cases. Even then, it swerves back and forth on the road whenever there were straight cracks on the road. It's good for 1k, but it's laughable you think it's as good as auto pilot in any way and definitely not worth acquiring. I'm not sure what benefit they'd gain over the current system.
Nice test/comparison,thanks for doing this and showing that it’s not so easy making a full self driving vehicle👍
Thanks for watching!
Openpilot is great. I ended up buying a 2020 corolla for this purpose and I’m not the kind of person who likes buying new cars. I regret not getting it sooner Openpilot is priceless and it should be worth more.
Great to hear!
WOW! One of the best objective evaluations of auto pilot I have ever seen. I have two Tesla s with FSD, but openpilot is very impressive for how it well it does for the cost and simplicity of hardware requirements. I was never going to buy a an internal combustion car ever again, but Openpilot is an effort that I want to support. am planning to buy a plug in hybrid Toyota RAV 4 just to support the amazing work that Comma AI is doing.
Great test guys, I didn't know Openpilot existed. Many legacy manufacturers should adopt this out of the box, than re invent the wheel!
You don’t have to have your hand on it the whole time. You can take your hands off and when the blue signal to touch it, you can just touch the volume button up or down and it goes away. So it really just wants to make sure you’re paying attention.
Deserves 0 dislikes. Good stuff
Thanks!
30 Tesla employees don't like it? lol
Doesn’t make sense to dislike. Both companies win from it
MrRamirez ToYou Definitely. I have that same Toyota Hatchback. I’m going to get it.
f0t0b0y have fun
Andy, At 9:20 when trying to initiate a lane change and you had to do it manually, it may have been due to an illegal lane change on a solid white line being prevented. I know everyone does it, but police will issue a ticket if crossing over a solid white line.
Yes it most likely was because of that
Great comparison video, loved it! It definitely cleared most of my concerns/questions regarding the Openpilot system. Thank you Andy and Logan!
Thanks for watching!
Solid and objective comparison between the two systems.
Well done.
You two are great collaborators!
VERY impressed with Openpilot. As a Model 3 owner with FSD Autopilot, I was expecting OP to be a big fail. It wasn’t. Great video.
Thanks for watching!
Ditto, except I have MS.
This was a great head-to-head real world test. I learned so much about the capabilities of both platforms as at 13 Jan 2020. Really enlightening. So grateful!
Thanks for watching!
"It's essentially a cellphone" I mean it literally is one lol. It's a OnePlus 3T doing all that heavy lifting.
Yes really impressive
@@aslye what if the phone glitches/ freezes?
@@howerover in Tesla they claim their Autopilot have 2 system running in parallel for redundancy, this is what OpenPilot should also try in my opinion..
How do you know its a Oneplus 3T? Is it stated is does it just have similar specs? Cool and weird shit either way.
@@davidcazares7441 because that's literally what they announced when they released the requirements to get OpenPilot up and running. They've also added support for the LeEco Le Pro3.
Openpilot is impressive. I have a 2013 ModelS without AP and thinkin about getting Openpilot. Can u do another vid during rain or snow with Openpilot?
Do you know why Openpilot decided to only go with one camera? This seems to cause most of the problems with it, and if its a cost thing, I'd be sure many users would pay an extra couple hundred dollars as a bonus upgrade for safety.
Mainly to focus on developing the software stack and trying to do as much as possible in building a good machine learning model for driving. In the future they may expand to more cameras but for now it's a 13 employee startup working very hard to improve what they currently have software wise! So much left to do with just single camera systems!
18:14 -- as far as I know when using OpenPilot if you press the accelerator then it will *disable* the 'self-driving' system. In Tesla Autopilot 'self-driving' system you can press the accelerator and the system stays active.
Openpilot disengages on acceleration or braking; it does not disengage if the driver turns the wheel. Autopilot disengages on steering or braking but not acceleration.
@@aslye , I was pointing out the difference with accel. Certainly, both (and all) systems would disengage with braking just like basic cruise control has done for decades. Very good point about Openpilot not disengaging when the driver turns the wheel while Tesla Autopilot would disengage AP(Autosteer) but leave AP(TACC) engaged.
Again better than lidar !
I don't know waymo seems to be doing it even better with lidar.
@@DerekSmit
Except they can't run their car on a road they haven't 100% mapped out. The way they are operating is completely unrealistic, and I don't know when they will stop working with fully mapped out roads and move to a dynamic model.
It's almost as if having lidar AND cameras is the way to go...
@@kendokaaa Nah, if you can do it with cameras, which is more difficult to achieve, you don't need lidar. Lidar is a crutch which has a ceiling, doing it entirely with cameras is the ultimate in cost effectiveness.
@Clutch Solution so you're claiming that cameras can see better in low visibility conditions than lidar, interesting. I didn't know cameras could see at a good distance in darkness and/or rain
Regardless of how well they both work, if you have to keep your eyes on the road, then you might as well drive yourself. Personally, I'd much rather have a HUD that flags potential threats in a non-intrusive way. This can be used to collect massive amounts of study cases for machine learning too.
I agree. I test drove a Tesla and regardless of how good it was at the time, it felt a lot like letting a learner driver drive. It should work well but you still have to keep your attention in case it doesn't and that I find more stressful than driving my self.
I find it takes more effort from my part because I have to look further ahead than I normally would then watch what it is doing and then try to predict if it will do the right thing. When I drive all thus happens automatically. I do not need to think or double guess what the system us going to do.
In my opinion when I do not need to pay attention and it will feel like a competent driver I will not use it. For highway cruising or slow traffic I think it will be ok.
Honestly it looks like if OpenPilot starts using multiple cameras it'll easily surpass Tesla autopilot in every scenario. I'm surprised that I've never seen much hype around such a great utility.
I tell people about it all the time, and none of them seem as excited about it as I am.
true, but tesla has over a billion miles of autopilot footage to play with. openpilot does not, aswell as the safety features
@@TypicalBlox So far OpenPilot has over 10 million miles. And it definitely has safety features lol.
@@outcast566 so will openpilot speed up if it spots a car zooming at it to the rear? th-cam.com/video/h3Dzi0_PmmU/w-d-xo.html plus 10 million miles is a 1/100 of tesla, not that impressive compared to over companies like google waymo
@@TypicalBlox AS it is now Tesla is better. But we can't overlook the fact that Tesla cars are factory made with autopilot compatibility. They use multiple cameras and radar. OpenPilot is using just one camera. Give it access to that many hardware and I don't think it'll be hard for Openpilot to match Tesla autopilot.
wow, what a great video, Watched all 30 mins of it. At least now we can see what owning a autopilot car is like. Maybe someday we'll be able to buy one. Thanks guys for the upload.
Thanks for watching!
Would be interesting see to how openpilot and autopilot handle driving in the dark, evening.
From what I hear it works great. These new cars have these super bright LED headlights that look like fucking high beams blinding oncoming traffic in the night lol. So OpenPilot should see just fine.
This was a very informative video on the progression of autopilot for not just Tesla but, a Toyota as well.
Thanks for the video!
I haven't been able to drive for years because of vision issues. Nice to see the current state of software-driven cars, especially the open-source solution. When I can get a car to drive me from point A to point B that will finally be something. Until then, I really hope some bad actor doesn't find a way to send their own OTA updates to people's cars.
Thanks for making this. It was very interesting to see a side by side compare that was done so well.
Thanks for watching, Michael!
Tesla following distance is not “car lengths”, btw; I believe it is also time-based (e.g., sensitive to speed). At setting 7 (max) it is a bit more than 2 seconds, which is the recommended distance.
Yes that is correct
Never heard of open pilot before this so this was super interesting to see