AI Learns Parallel Parking - Deep Reinforcement Learning

Subscribe for more content like this:

Follow me on Twitter for more frequent updates on my projects:

Basically, the input of the Neural Network are the readings of eight depth sensors, the car's current speed and position, as well as its relative position to the target. The outputs of the Neural Network are interpreted as engine force, braking force and turning force. These outputs can be seen at the top right corner of the zoomed out camera shots.

The AI starts off with random behaviour, i.e. the Neural Network is initialized with random weights. It then gradually learns to solve the task by reacting to environment feedback accordingly. The environment tells the AI whether it is doing good or bad with positive or negative reward signals.

The training was done on a computer with an i5 (7th or 8th gen) and a GTX 1070 with 100x simulation speed, using 6 instances of the environment and up to 6 processes running in parallel.

Timelapse Music: "The Elevator Bossa Nova"
Outro: "All That"

#ArtificialIntelligence #MachineLearning #ReinforcementLearning #AI #NeuralNetworks #hostinger #inspeedwebelieve #speedfreak

[SamuelArzt]

If you're looking to create your own website, I genuinely recommend this video's sponsor, Hostinger.

[ProjectPhysX]

And now try it with two cars competing for a single parallel parking lot. I want to see some reckless AI driving!

[jeI]

i'm a high school student in Korea who asked for your advice last time. Thanks to your help, it was solved well last time. Thank you. Whenever I see your research, I am always amazed and want to emulate you.
I'll always support you in Korea🇰🇷🇰🇷🇰🇷

[ROBA]

1:58 "I'll just drive home"

[HakikatBirSinghBhatti]

Saw this video last year, had a wonderful ride through the youtube algorithm, came across the 2 min papers yt channel. I'm a doctor by training, completed my MBBS last yr but now pursuing AI, ML and Data Science Post Graduation program for it's applications in Healthcare.

So, Yes. YOU ARE MY INSPIRATION, Sir.

[OfficialYunas]

Increasing the environment complexity while training usually helps. For example, start with the cars far apart and a huge desired parking space. Then gradually make the environment harder while training, e.g. cars less apart and parking space smaller.

[revimfadli]

Big yes to the format, informative & shows the thought process.

Btw would hindsight experience replay help with the learning?

[Firzj]

where are the new videos? Did the AI overtake your channel ? 👍👽

[Kram]

try something like a slow but steady increase in difficulty wherein at first just getting close is enough and then gradually the difficulty increases until only near-perfect parking is possible.
You could even find inspiration in POET (Paired Open-Ended Trailblazer) and its recent improvement Enhanced POET wherein, basicaly, the concept is that the environment gets its own process (they used evolution but you could also make it some learning algorithm) to gradually modify the environment in ways that try to keep the agent as close to the frontier of its abilities as possible, allowing for gradual improvement.
I suspect that'd be a lot of work though. It'd involve some sort of controllable procedural generation of your city scene. Currently that, as far as I can tell, is lovingly handcrafted.

[SpEsHuL]

I wish this guy made more AI Content especially with the new advancements in AI

[berkayherek]

To be honest I think it was really good video and nice explanations I liked how you change stufs and tell us everything.

[MudakTheMultiplier]

I know it's been a year, but I feel like scaling the perfect parking reward exponentially would help guide the ai into actually finding the spot.

[algorithminc.]

Fun channel. Subscribed. Who needs comedy shows, when they can watch AI systems being trained. Always a good laugh. Reminds me of back in the 80's and 90's, the interesting results of textual systems - the early attempts of AI. There's some of that in the interesting five-part series "The Machine that Changed the World." I have some of those old-school videos in the "business" section of my YouTube, if you have interest. Thanks for the great channel ... enjoying it.

[MrCool-lols]

Hey, I've had this idea now for quite some time but I lack the experience and capabilities to try it out myself:
You could probably have the ai begin learning in an easy environment and once you're happy with the results make the rules a bit more strict - kinda like school but the ai is only judged not stressed. For example the the ai might learn to do simple stuff like get close to the parking spot, next you make rules more strict for the position, then it has to face the right way and last, it needs to actually stopand after that you could add all kinds of fancy rules to a somewhat trained ai.
I've thought about a similar thing with the network itself, small networks are fast to train but have not much complexity and big ones take an eternity to train. So what about starting with a small network and having it grow whenever there seems to be no improvement on tge results?

I would absolutely love these concepts to be tried out but I just cant do it myself yet.

[kiwi__official]

will you make more of these? it would be nice to see AIs try to drive on roads and go from point A to point B or go through courses as fast as possible or some things like that

[komi]

can you teach us how to make a basic ai that does stuff like this please! I'm really interested in how you get this to work.

[robertjames]

Here's an idea for your next video. Make the car learn how to drive and park at a specific location and it needs to stay in the correct lane of traffic

[oxiigen]

great and interesting research! thank you!

[Keamybakunin]

Hello is it possible to get the resources used in the video ? The parking and the vehicule ? Thanks.

[mbeware]

It would be interesting to see the AI select the best parking spot... And I don't know if there is a time rewards (less time bigger reward). And for the strange parking methods of entering the parking spot to reverse, maybe give a penalty where going on the curb??