“We model the unit to drive like the drivers of that vehicle normally drive.” So, no turn signals when installed in BMWs, and constant turn signal for 15miles when installed in MB, and doing 40 mph in the passing lane for Priuses?
The great thing about this being open source is that you can fork it on github and make the aptly named _is_driver_distracted function always return false, recompile, and then watch Harry Potter on DVD like that guy who wrecked his Tesla.
Pie 4life he didn’t promote it, the software just didn’t check for it. Currently nothing prevents you from looking at your phone so it’s better than nothingz
@@Pie4life123 actually he said the exact opposite the only time he said it was fine to look at your phone was when the car was completely stopped then you have to look up when it goes and it watches your face to make sure you are looking straight ahead, I think you should re watch the video
@@MrPacMan36 If it's only using the cameras on the modified phone it will never be as good as Autopilot for the simple reason that it just cannot "know" as much. For example, if there's a car on your right that is not visible through the front window yet and he's moving into your lane, this system cannot do anything about it because it doesn't see that car. Autopilot, on the other hand, has numerous cameras and radar systems to get that info and adjust accordingly. Same goes for a car coming up behind you really fast. This system would not even know it was there, but Autopilot would and adjust accordingly. I'm sure it will get better as time goes on, but I hope the next iteration includes additional cameras for the sides and back of the car.
Nice! Hope they can clean this up for a more OEM look without the exposed wires and reduce the number of steps to install. Notice how the car plastic panel behind the rearview mirror has to stay off, exposing a bunch of wires. Then there is a wire hanging, the soldering of the circuit, the software flash via the OBD port, etc. during the install process. All of that gives a feel that it's a less-than-polished solution.
George Hotz comment in the video are incorrect about Tesla AutoPilot not slowing down for turns. It does and there are many youtube videos showing it. It can see/calc how sharp turns are vs your current speed.
You can order one if you go to commaai website - shop. Make sure your car is compatible you can see all the cars supported on commaai github - openpilot.
Damn this system looks really promising although I do have an idea to potentially improve on the system. Change the timing for the auditory warning a bit or give the user the option set the limits stricter if he wants that and maybe think about including an automatically adapting system that changes the timing of the warnings based on traffic density and time of day.
Is it possible to use it only as a warning system for lane drifting, collision etc? I guess there is a barrier for most people to put their lives on a $500 device, whereas a $500 passive driving monitoring system is quite affordable for those who can't upgrade their old cars yet.
This system appears to work quite well, especially for the money, but if it is only using the cameras on the phone its usefulness will be limited. Can it see a car on your right about to turn into your lane and hit your car - and get out of the way - like Tesla Autopilot can? Or if a car is coming up behind you really fast? Fantastic technology, and better than not having anything to augment the human driver's input, but I hope the next iteration includes additional cameras.
Very interesting. I recognize the route: South bound on S.F.'s Portola/Junipero Serra Blvd, 280 South, then looping back North on Junipero Serra from South S.F. That's a decent drive, but I can't help but wonder how the system would work during commute times for the S.F. to Silicon Valley commuter.
Writing a algorithm for a self-learning car is not that difficult. As the name suggests, its "self-learning". All you have todo is describe your model and setup your network. (That can be done in under 200lines of code actually). To train the model you would use data, in this case visual data, converted to a 1D Array and the apropriate angle of the steering wheel. If you feed the System with enough data, it will learn how to drive! Now a little Q and A: Q: Why wasn't that done before if thats so easy? A: It was done before, WAY in the past actually, here is an example after a quick google search: www.theverge.com/2016/11/27/13752344/alvinn-self-driving-car-1989-cmu-navlab Q: Why are you of the opinion, that this is not save? A: The Neural Network, or AI as you would call it is not the problem here. The problem is the insuficient hardware. It is not enough to train the described model only by visual data. Surely it will work (as seen in the video), but it will fail as soon as the image is disturbed by (some examples): - Mud on the camera lense -Mud on the Windshield -Bad Weather -Weak illumination (eg. at night) - A software crash on the phone Q: Why do you think, that this concept will never work? A: In order to react to the traffic, the car needs only one thing, but a lot of that. And that thing is data. Teslas and Googles cars work, because they have dozens of sensors in them to provide the AI with a lot of various information, to act upon. Whereas the concept from the video only uses non-redundant visual data and only from a narrow perspective instead of an 360 degree image
You have clearly little idea of how complex machine learning is. You can't make a complete machine learning progress like this one with 200 lines of code. And if that's so easy why don't you build your own self driving car and release it, this guy got quite a lot of money from this startup
This guy is a scammer but "Writing a algorithm for a self-learning car is not that difficult" is just total bullshit. The great challenge of autonomous driving is machine vision / object recognition, not a lack of sensors. You can put any sensors whatever on a car if you want.
no its not. Its nothing more than a students project in a machine learning course at university. It is hard to bear, how he can fool so many people with this little tech demo.
Cheap isn't cheap if you accidentally blow out the original electronics. Oops, a transistor just fried. Oh, that can never happen. OK, then give a written guarantee for the cost of repair or total car replacement. You must have not followed instructions. Sorry, in that case, we don't pay.
5:33 There you have it. "Because it's hard to do". This guy is selling you already common technologies and he has no clue how to even begin to think about the real problems in autonomous driving. Seriously, have you guys never heard about ACC and lane keeping assist?
Okay, so how does that make this bad? The software aspect has already been done, but to make the hardware this minimal and effective while being < 1000 dollars has not. "No clue how to even begin to think about the real problems in autonomous driving" -- Why are you so insulting? I can guarantee you he has thought about this problem a lot more than you have.
This only works on cars that have sensors and cameras, it replaces the ACC and LKA, with a system that is open source and better (but clearly not self-driving). He's not inventing some brand new technology, he's creating a platform that takes driver assisted cars and makes them smarter.
The difference is that the people working for the companies may not have the interest in making the best system. Instead, they just want to pay their bills. This guy on the other hand seems to love what he does and this is why his system is going to be the best.
You do realize he is also a genius the first person to jailbreak an iphone and ps3 also he founded a company that does it. It isn't just a guy in his apartment and they have thousands of ppl using this system and it is always learning. and yes big companies can do the same thing level 2 is available on autopilot and super cruise.
Jason, that is the responsibility of the government to figure out, but those vehicles are not safe and can even cause people to be less attentive. One suggestion on how to handle them is to have roads like car share lanes. They could turn a 4 lane into a three lane with use one lane for the autonomous vehicles. Regardless how they do it they have no business driving with non autonomous vehicles.
That's just not possible, they have to find a middleground there and the driver needs to be reminded to keep there eyes on the road, I mean you could be distracted driving your car as well, just stare at your infotainment system for too long and voila, before you know it you have not looked on the road for 2-3 seconds, maybe more, but I do think that they should change the timing for the warning systems a bit, because I think 4 Seconds are too much, make it maybe make it 3.5 or 3 for the audio warning or at least give the option or make an option where it adaps automatically depending on how much traffic there is and what time of day it is as well.
John Smith ok, then how do you think that's going to work? Build new streets, way to expensive, also what about Citys where do you wanna build These lanes, go ahead tell me how you would do it in a realistic, financially feasable way.
homeboy is genuis. first to hack playstation, unlock iphone and this. keep it up.
He never managed to hack the PS3.
@@HaploPrime Then why was Sony making his life difficult?
@@consig1iere294 Because he hacked the PS3.
“We model the unit to drive like the drivers of that vehicle normally drive.”
So, no turn signals when installed in BMWs, and constant turn signal for 15miles when installed in MB, and doing 40 mph in the passing lane for Priuses?
You can get the finger flipper for 199$ extra.
Won't be surprised if there will be joke car plugins like these.
The great thing about this being open source is that you can fork it on github and make the aptly named _is_driver_distracted function always return false, recompile, and then watch Harry Potter on DVD like that guy who wrecked his Tesla.
Amazing, lol.
perfect.
🤣🤣🤣
I'm usually anti-autonimous driving, but this is tech I can get behind if it means everyone else can drive safely for cheap.
Thai Win wrong, this dumb ass promote texting and driving literally the number one cause of traffic deaths and accidents if distracted driving.
Pie 4life he didn’t promote it, the software just didn’t check for it. Currently nothing prevents you from looking at your phone so it’s better than nothingz
he said it also watches your head , if you look away for 2 secs or more it will cause an alert the escalates to safely slowing the car down
@@Pie4life123 actually he said the exact opposite the only time he said it was fine to look at your phone was when the car was completely stopped then you have to look up when it goes and it watches your face to make sure you are looking straight ahead, I think you should re watch the video
Cheap anything is not good.. Car can veer off on it's own. and bam....
is this the same guy who had an email argument with Musk over who can make a better self-driving car system and said was going to make a company?
yup
Yup and this is his system
Yup. And this guy's system works safer and smoother than Elons system. Just needs to gather more data and get out there
@@MrPacMan36 If it's only using the cameras on the modified phone it will never be as good as Autopilot for the simple reason that it just cannot "know" as much. For example, if there's a car on your right that is not visible through the front window yet and he's moving into your lane, this system cannot do anything about it because it doesn't see that car. Autopilot, on the other hand, has numerous cameras and radar systems to get that info and adjust accordingly. Same goes for a car coming up behind you really fast. This system would not even know it was there, but Autopilot would and adjust accordingly.
I'm sure it will get better as time goes on, but I hope the next iteration includes additional cameras for the sides and back of the car.
@@JustWasted3HoursHere it uses the car sensors as well as the camera data
I'm assuming installing this in a Yugo GV Plus WITH the Automatic of course, wouldn't be an issue?
Nice! Hope they can clean this up for a more OEM look without the exposed wires and reduce the number of steps to install. Notice how the car plastic panel behind the rearview mirror has to stay off, exposing a bunch of wires. Then there is a wire hanging, the soldering of the circuit, the software flash via the OBD port, etc. during the install process. All of that gives a feel that it's a less-than-polished solution.
Getting my dev kit in the mail today! Open source installing this on my Pre Auto Pilot Tesla 😀
Well good for his open system. The retro fitting ability will make him HUGE money.
George Hotz comment in the video are incorrect about Tesla AutoPilot not slowing down for turns. It does and there are many youtube videos showing it. It can see/calc how sharp turns are vs your current speed.
Were can I buy one ?
You can order one if you go to commaai website - shop. Make sure your car is compatible you can see all the cars supported on commaai github - openpilot.
So, how long before I can install this in my Yugo? (:
Damn this system looks really promising although I do have an idea to potentially improve on the system.
Change the timing for the auditory warning a bit or give the user the option set the limits stricter if he wants that and maybe think about including an automatically adapting system that changes the timing of the warnings based on traffic density and time of day.
How are you mounting the camera to the roof?
Is it possible to use it only as a warning system for lane drifting, collision etc? I guess there is a barrier for most people to put their lives on a $500 device, whereas a $500 passive driving monitoring system is quite affordable for those who can't upgrade their old cars yet.
won't work. This system only works on newer cars that have adaptive cruise control and lane departure warning.
I want one... 😂. Looks great to me....
This system appears to work quite well, especially for the money, but if it is only using the cameras on the phone its usefulness will be limited. Can it see a car on your right about to turn into your lane and hit your car - and get out of the way - like Tesla Autopilot can? Or if a car is coming up behind you really fast? Fantastic technology, and better than not having anything to augment the human driver's input, but I hope the next iteration includes additional cameras.
Will this work in a 2014 Accord Hybrid EX-L? It does have electronic power steering.
Very interesting. I recognize the route: South bound on S.F.'s Portola/Junipero Serra Blvd, 280 South, then looping back North on Junipero Serra from South S.F. That's a decent drive, but I can't help but wonder how the system would work during commute times for the S.F. to Silicon Valley commuter.
Wow, that guy was nervous.
BTW: In some States, (CA), you can't mount stuff on your dash.
Dave Taylor seems like his head was pinned to the headrest.
I can seriously see car manufacturers including this in their cars in future. It can't happen soon enough really.
Emme I like this new ride along style review.
R Collinge if only her passenger made more human-like expression face things.
I have a 2018 accord touring with safety features standard... should I still get this? Anyone in same senário?
Get it. You won't regret it.
people on github and stackoverflow are soo good that they are going to make it the best like linux
His uh-huh sounds like Mickey Mouse
I have a Prius Prime. What do I need to get that working with this
Money! XD
The answer is fucking yes.
Tesla's Autopilot DOES slow down for curves.
Can someone explain how he got a regular car to get its steering wheel to automate ?
They are cars with assisted driving already, and they tap into that system and send it new commands. This is why only certain cars are supported.
such a great idea!
Just take my money!
At 2:05 i thought he makes the sign of cross lol
Um.. Tesla's Autopilot DOES slow down for curves.
technically this does as well by no longer using the gas it will start to slow down it just doesn't brake in advance of turns to slow down
that's because of the regenerative breaking in the tesla when you let go of the gas
Writing a algorithm for a self-learning car is not that difficult. As the name suggests, its "self-learning". All you have todo is describe your model and setup your network.
(That can be done in under 200lines of code actually). To train the model you would use data, in this case visual data, converted to a 1D Array and the apropriate angle of the steering wheel. If you feed the System with enough data, it will learn how to drive!
Now a little Q and A:
Q: Why wasn't that done before if thats so easy?
A: It was done before, WAY in the past actually, here is an example after a quick google search:
www.theverge.com/2016/11/27/13752344/alvinn-self-driving-car-1989-cmu-navlab
Q: Why are you of the opinion, that this is not save?
A: The Neural Network, or AI as you would call it is not the problem here. The problem is the insuficient hardware. It is not enough to train the described model only by visual data.
Surely it will work (as seen in the video), but it will fail as soon as the image is disturbed by (some examples):
- Mud on the camera lense
-Mud on the Windshield
-Bad Weather
-Weak illumination (eg. at night)
- A software crash on the phone
Q: Why do you think, that this concept will never work?
A: In order to react to the traffic, the car needs only one thing, but a lot of that. And that thing is data.
Teslas and Googles cars work, because they have dozens of sensors in them to provide the AI with a lot of various information, to act upon.
Whereas the concept from the video only uses non-redundant visual data and only from a narrow perspective instead of an 360 degree image
You have clearly little idea of how complex machine learning is.
You can't make a complete machine learning progress like this one with 200 lines of code.
And if that's so easy why don't you build your own self driving car and release it, this guy got quite a lot of money from this startup
Also that self driving car you've shown almost fell of a simple turn going at 15 km/h.
It's no where near the things shown in the video
This guy is a scammer but "Writing a algorithm for a self-learning car is not that difficult" is just total bullshit. The great challenge of autonomous driving is machine vision / object recognition, not a lack of sensors. You can put any sensors whatever on a car if you want.
Have worked on a self driving car, I can tell you this is dead wrong. The problem is the algorithms. Not the hardware.
How can she look like a grandma and a teenager at the same time
Is open pilot able to recognize adults, kids, animals, baby strollers, and avoid hitting them?
I'm still waiting for a car which I can take 20 min rest while it drives on the freeway itself...
Geohot is the fuckin man god damn
Damn, this guy was under-rated; he’s a genius and has brilliant tech. Open-source? Wow awesome, this is more the future than Tesla.
no its not. Its nothing more than a students project in a machine learning course at university. It is hard to bear, how he can fool so many people with this little tech demo.
@@someguy5049 not even Havard computer science student could do something like this, it's far more complex than you think
This is an old video
I would not put my life in the hands of a car. Thats all.
can i use it for toyota yaris 2003? manual gearbox
Cheap isn't cheap if you accidentally blow out the original electronics. Oops, a transistor just fried. Oh, that can never happen. OK, then give a written guarantee for the cost of repair or total car replacement. You must have not followed instructions. Sorry, in that case, we don't pay.
He is litle..... hm
zuti071 He is an engineer. A pretty talkative one. But yeah, we all have Asperger's to some degree
In short, he is awkward. You are obviously his lawyer and now i have new mystery. What tf is Asperger and how is that thing in me ?
"A little..."? He is plenty to me...
Dude is a kind of a weirdo.
I heard it's nasty to work with him
Nope he’s a genius and humble guy
i drive 2018 prius with toyota safety sense plus safety suite. i don't need this...
Well then fuck off
5:33
There you have it. "Because it's hard to do". This guy is selling you already common technologies and he has no clue how to even begin to think about the real problems in autonomous driving.
Seriously, have you guys never heard about ACC and lane keeping assist?
Okay, so how does that make this bad? The software aspect has already been done, but to make the hardware this minimal and effective while being < 1000 dollars has not. "No clue how to even begin to think about the real problems in autonomous driving" -- Why are you so insulting? I can guarantee you he has thought about this problem a lot more than you have.
This only works on cars that have sensors and cameras, it replaces the ACC and LKA, with a system that is open source and better (but clearly not self-driving). He's not inventing some brand new technology, he's creating a platform that takes driver assisted cars and makes them smarter.
Billon dollar company’s can’t perfect this so your telling me a guy who lives in his apartment made his own? It’s not safe
It's an open source system. He didn't develop the entirety of it by himself.
The difference is that the people working for the companies may not have the interest in making the best system. Instead, they just want to pay their bills. This guy on the other hand seems to love what he does and this is why his system is going to be the best.
You do realize he is also a genius the first person to jailbreak an iphone and ps3 also he founded a company that does it. It isn't just a guy in his apartment and they have thousands of ppl using this system and it is always learning. and yes big companies can do the same thing level 2 is available on autopilot and super cruise.
Can't perfect what? What did you actually see in the video?
These should NOT be allowed on the same roads with non autonomous vehicles. They should have their own roads.
John Smith lol where are these roads going to come out of?
Jason, that is the responsibility of the government to figure out, but those vehicles are not safe and can even cause people to be less attentive. One suggestion on how to handle them is to have roads like car share lanes. They could turn a 4 lane into a three lane with use one lane for the autonomous vehicles. Regardless how they do it they have no business driving with non autonomous vehicles.
That's just not possible, they have to find a middleground there and the driver needs to be reminded to keep there eyes on the road, I mean you could be distracted driving your car as well, just stare at your infotainment system for too long and voila, before you know it you have not looked on the road for 2-3 seconds, maybe more, but I do think that they should change the timing for the warning systems a bit, because I think 4 Seconds are too much, make it maybe make it 3.5 or 3 for the audio warning or at least give the option or make an option where it adaps automatically depending on how much traffic there is and what time of day it is as well.
Last, yes that is VERY possible. autonomous vehicles have no business on the same roads as non autonomous vehicles
John Smith ok, then how do you think that's going to work? Build new streets, way to expensive, also what about Citys where do you wanna build These lanes, go ahead tell me how you would do it in a realistic, financially feasable way.
Well good for his open system. The retro fitting ability will make him HUGE money.
Getting my dev kit in the mail today! Open source installing this on my Pre Auto Pilot Tesla 😀