"reinforcement learning does not need a big set of training data". This is not true. RL requires a much bigger sample complexity than supervised learning.
I've written an AI for a racing game before, I'd be curious to see how long it would take to train an AI to drive as well as the purpose built AI I wrote or even if it could drive that well at all (seeing how they drove in the video did not inspire confidence in machine learning).
I think the difference is that when you're writing an AI for a game, the AI has full knowledge of the track and locations of other vehicles and objects. For a machine learning AI, it only has the input from its sensors, so it has to be able to generalize and make decisions on the fly.
@@Lloyd2112DT I got the impression that they were abstracting that part of it though as they mentioned you would get distance from the center of the road, and some code looked like it showed stuff like which wheels were off track so the inputs would be pretty similar to what I was working with.
@@AusSkiller Nah. I've done a lot of computer vision/self driving projects similar to this. You generally only get the camera view from the front of the car (if you're lucky, there are additional cameras and LIDAR), and you need to train the neural network to detect the edges of the track and write a PID controller to try to stay within the boundaries without swerving too much. In the video they also showed a 3D simulation of the track, and that's generally used to generate additional training data and to prototype networks that are still in development because training on actual camera footage from an RC car is so much more expensive.
@@Lloyd2112DT Oh, so this whole thing is just a matter how well it identifies the road then? Well I'm completely uninterested in that. I don't want to be wasting time on having to re-implement yet another half assed computer vision library for the umpteenth time and then have to sit around training it before I can even get usable data for a practical application.
@@AusSkiller Yup. Trying to train a neural network to deal with that stuff seems pretty tedious to me as well. I guess that's why Amazon isn't paying us the big bucks? ¯\_(ツ)_/¯
Regrading deep learning, I think the xiaorgeek donkey car also can do that: www.xiaorgeek.net/collections/donkey-car/products/xiaor-geek-donkey-car-xr-f2-with-nvidia-jetson-nano-developer-kit
oh god, get that corporate hack that starts saying "in a fun and new way" every time you pull his string in the back out of here and get one of the real nerds that code that stuff and actually know what they are doing in here. on the other hand, I guess this "person" shows how could they have already gotten with speech synthesis.
And Deep Racer is not available out of the USA (at least from France). DonkeyCar seems interesting. If you are in Paris Area you can check IronCar as well.
We did the trials in 1996, pitting autonomous vehicles with kids from a local pre-school up the street from Oakland University in Michigan. onlinelibrary.wiley.com/doi/abs/10.1002/rob.20028 we discovered then that machine intelligence was around a 2-3 yr old, in 1996....article published in 2004.
Autonomous cars don't really learn, they use the code that has been written in them and act upon that code. Here we see the cars learn to drive and everything involved with it rather then just use prior inputs we have given to them Hope it clears it up :)
Google: We see where you go on the internet and sell you ads. Oh and we have our self driving cars Facebook: We see you, listen to you and buy all sorts of data on you to sell you more stuff. Oh we have our own VR system and money. Amazon: Screw the other guys, we are going to crowd source our Autonomous car software to software engineer. No wonder only one of the 3 company owner has a rocket company. Not hating on Amazon. I just finished a project with Blue Origin. If my wife would move with me, I'd be there in a heartbeat
It's funny how they call things machine learning today. When before it was just code. Code that recognizes something and takes note. Changes a variable and set. It's just simple automation and programming.
You ain't gonna believe what I just seen! Tyra Banks? See, now you've said something better, so my thing don't sound cool no more. Oh! Come come on, what'd you see? I don't remember. I just was thinking about Tyra Banks! Can you imagine? Every night! Don't shake my hand. You ain't got no hand! You're a little truck!
I don't think he needs much support. He could pay someone millions of dollars a day just to follow him around all day and support his fucking ballsack. Fuck billionaires, they need to be spanked by the government and forced to pay a shitload more taxes
The car running off the road in the background was gold
@Gōdon Gurando yes let's not let facts get in the way....fkn leftist
1:20 LMFAO Idk why I find this *_SOOO funny_*
That........ Was hilarious.
Like the car was curious "You guys talking about me?"
"What is machine learning?"
"It is not what the car behind me just did"
"Crash?"
"We don't use the word crash"
Photo bomb!!!
2019: Awwww so cute. They are learning.
2059: RUN! DeepRacer is coming! ARGGHHHHH
Chua Koktung Dead AF
Hahaha that was funny, the robotic car ran off the track as they were talking!
proportion steering will help the models be a bit more smooth making the turns
PID tuning is how I would write the ai Code
yeah when I first saw that I was like wtf
Finally after seeing AWS updates on Formula 1 for years, I gotta see what it really is!!!
"reinforcement learning does not need a big set of training data". This is not true. RL requires a much bigger sample complexity than supervised learning.
Lmao 2:51 computer aggression.
When the general manager started speaking, I was just like..... interesting, you know a lot.
I subscribed to Norm Chan's Tested because I love the videos that his occasional guest host Adam makes.
These cars are really funny to watch I love it
I've written an AI for a racing game before, I'd be curious to see how long it would take to train an AI to drive as well as the purpose built AI I wrote or even if it could drive that well at all (seeing how they drove in the video did not inspire confidence in machine learning).
I think the difference is that when you're writing an AI for a game, the AI has full knowledge of the track and locations of other vehicles and objects. For a machine learning AI, it only has the input from its sensors, so it has to be able to generalize and make decisions on the fly.
@@Lloyd2112DT I got the impression that they were abstracting that part of it though as they mentioned you would get distance from the center of the road, and some code looked like it showed stuff like which wheels were off track so the inputs would be pretty similar to what I was working with.
@@AusSkiller Nah. I've done a lot of computer vision/self driving projects similar to this. You generally only get the camera view from the front of the car (if you're lucky, there are additional cameras and LIDAR), and you need to train the neural network to detect the edges of the track and write a PID controller to try to stay within the boundaries without swerving too much.
In the video they also showed a 3D simulation of the track, and that's generally used to generate additional training data and to prototype networks that are still in development because training on actual camera footage from an RC car is so much more expensive.
@@Lloyd2112DT Oh, so this whole thing is just a matter how well it identifies the road then? Well I'm completely uninterested in that. I don't want to be wasting time on having to re-implement yet another half assed computer vision library for the umpteenth time and then have to sit around training it before I can even get usable data for a practical application.
@@AusSkiller Yup. Trying to train a neural network to deal with that stuff seems pretty tedious to me as well. I guess that's why Amazon isn't paying us the big bucks? ¯\_(ツ)_/¯
2:56 they used shingo as the training data
If you like this you may want to look for apha star videos. It's an AI playing on the official starcraft 2 ladders.
My fav right now
subnautica fans, I heard reinforcement and platform in the same scentence and thought "ENFORCEMENT PLATFORM
Wow, Amazing!
1:23 could use a few more generations lol
even humans can be confused 😁
So Cool
Cyber Dyne Systems skynet kept learning at a phenomenal rate until . . .
It became self aware !
#werefucked
So when do we start replacing drivers of real cars in real races with DeepRacer?
It triggers me the fact that they're going backwards...
This is fucken brilliant
Why don't they just show the Kiva line of bots they bought out and use as part of their operations? That's the smart bot they have already.
6:56 I think, I get the point of this research. 😄
Yes Amazon is entering self driving taxis.
How AWS Formula 1 predictions work???
Tested do that!!!!!!!
Regrading deep learning, I think the xiaorgeek donkey car also can do that: www.xiaorgeek.net/collections/donkey-car/products/xiaor-geek-donkey-car-xr-f2-with-nvidia-jetson-nano-developer-kit
AI is about to hit a wall we need QC ASAP
oh god, get that corporate hack that starts saying "in a fun and new way" every time you pull his string in the back out of here and get one of the real nerds that code that stuff and actually know what they are doing in here.
on the other hand, I guess this "person" shows how could they have already gotten with speech synthesis.
I've seen this before. Duplicate mailing.
First time I've seen Tested make this error.
FYI: DonkeyCar is an open-source alternative, with a deep community, which is far ahead of DeepRacer. watch?v=MEtfUqxjPmI
And Deep Racer is not available out of the USA (at least from France). DonkeyCar seems interesting. If you are in Paris Area you can check IronCar as well.
why is it built like its going in reverse lol
😂😂😂
deepracer.ai racer.ai races.ai sport.ai
Disappointed you aren’t being more critical of Amazon. Kind’ve deifying it, actually.
Did you mean critical?
Matthew B. Thank you, I did.
what is he supposed to say? bring activistic questions to an engineer for deepracer? That's not what this video is about
We did the trials in 1996, pitting autonomous vehicles with kids from a local pre-school up the street from Oakland University in Michigan. onlinelibrary.wiley.com/doi/abs/10.1002/rob.20028 we discovered then that machine intelligence was around a 2-3 yr old, in 1996....article published in 2004.
Amazon pays people to play with toy cars. Good to know
We have autonomous cars. How is this more interesting?
Autonomous cars don't really learn, they use the code that has been written in them and act upon that code.
Here we see the cars learn to drive and everything involved with it rather then just use prior inputs we have given to them
Hope it clears it up :)
Google: We see where you go on the internet and sell you ads. Oh and we have our self driving cars
Facebook: We see you, listen to you and buy all sorts of data on you to sell you more stuff. Oh we have our own VR system and money.
Amazon: Screw the other guys, we are going to crowd source our Autonomous car software to software engineer.
No wonder only one of the 3 company owner has a rocket company.
Not hating on Amazon. I just finished a project with Blue Origin. If my wife would move with me, I'd be there in a heartbeat
It's funny how they call things machine learning today. When before it was just code. Code that recognizes something and takes note. Changes a variable and set. It's just simple automation and programming.
They've always called this kind of programming machine learning? Not all code is machine learning. But all machine learning is code
Hey norm from tested here *me exits
I am very frightened that this shitshow looks great for full scale.
You ain't gonna believe what I just seen!
Tyra Banks?
See, now you've said something better, so my thing don't sound cool no more.
Oh! Come come on, what'd you see?
I don't remember. I just was thinking about Tyra Banks!
Can you imagine?
Every night! Don't shake my hand.
You ain't got no hand! You're a little truck!
Someone call Sarah Connor. We've for them.
Still waiting for a non BS answer to why Amazon is doing this. Shady answers from a shady company
I want to upvote but it is illegal to support Bezos publicly.
Jttv Solidarity Comrade
I don't think he needs much support. He could pay someone millions of dollars a day just to follow him around all day and support his fucking ballsack. Fuck billionaires, they need to be spanked by the government and forced to pay a shitload more taxes
Anti yells in the corner
That's a lot of bs to make a car go round a track, no wonder their so crap.
3rd
Holy shit their AI is HORRIBLE.
sounds like rats in the background
This will ruin the world and all of our lives...ain't that just cute
FiRSt commento
First