Automated Parking Using RL, a Unity ML-Agents Tutorial.

Подписаться 113

Просмотров 13 тыс.

50% 1

In this tutorial I will show you, how to use ML-Agents, and create an automated parking system using reinforcement learning.
Intro: 0:00
Requirements: 1:03
Installing Unity / Opening the project: 1:36
Installing ML-Agents (Unity): 2:49
Installing Python / ML-Agents: 4:02
Installing PyTorch: 6:16
Verifying the install: 7:09
Taking a look at the environment: 7:46
Taking a look at the agent's code: 18:29
The hardest part, rewards: 21:17
Preparing the environment for training: 22:26
Creating a demo file: 24:13
Looking at the trainer config: 25:43
Training from the editor: 31:45
Spinning up Tensorboard: 33:23
Training from a build: 35:38
Checking statistics: 38:26
Deploying the model: 39:25
Github: github.com/VanIseghemThomas/A...
My website: www.thomasvaniseghem.be/portf...
Proximal Policy Optimisation: openai.com/blog/openai-baseli...

Опубликовано:

26 июл 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 33

@maciejjasielewicz9607 2 года назад

Thank you for sharing, your video helped me big time!

@spulberioanalexandru5180 Год назад

Im new to this NN-Unity projects and you really explained the logic behind very well and did a very nice project.

@wendyqiu2692 2 года назад

Great work! Thank you for sharing!

@Diego0wnz 2 года назад

Nice one!

@UhGoomba 2 года назад

Great stuff

@ThalisUmobi 2 года назад

Real people use the other car bumper exploit in real life all the time too! LOL. Great video mate! Keep it up!

@thomasvaniseghem 2 года назад

Haha thank you!

@erenyeager4502 2 года назад

Thanks for this

@cassioj2 2 года назад

THANKS I LOVE YOU

@marcinziajkowski3870 6 месяцев назад

13:40 Well why do we set detectable "tags" in Ray Perception Sensor 3D if those rays can see only distance? How to make them recognize object types/tags/layers ?

@spidermanclips7966 Год назад

Hi there , do you have models for parallel parking too ? can you share it if possible ?

@marcinziajkowski3870 6 месяцев назад

How can I lookpu/preview data that are sent from ray 3d to Observer ?

@mateusdsp6871 5 месяцев назад

Goodd! :D Jesus is coming back, God bless!

@baselmohamed2767 Год назад

Hey is it a problem that it doesn't show the unity logo in cmd. I followed the same exact steps but that is the only thing not showing. Thank you btw

@Tehunknowntwin Год назад

I can't really understand how this is working? Your only observations are the speed of the car and the ray perception sensors. Yes the car gets point for going in the right direction, but it doesn't know the position/rotation of itself and the target so I am wondering how it can make sense of anything? Could you elaborate :) I have been trying to get my own car to park for a long time now, and your example seems to be working, but I just can't grasp why it does.

@yebv Год назад

That's really a nice project, but I want to know if I want to train this env by other algorithms, how can I do that? Thank you!

@thomasvaniseghem Год назад

You would need to edit the trainer config to use another suported algoritm such as Soft Actor Critic. I tried to get this working but I couldn't in time and haven't really found an oppertunity to look at it again. Another way is by connecting it to your custom Python instance and handle it as a gym environment. An example notebook for this can be found in the repo!

@WAMBHSE 2 года назад

I'm having real trouble finding good video tutorials on how to use curriculum training. Could you please make a video on this topic. Show a really simple example? Thanks.

@thomasvaniseghem 2 года назад

That is one of the topics I'm planning to look in to after it being suggested by someone already! I'll see if I find the time to make a demo and maybe a video about it. It will take a little while though.

@GetBetterNBetter 7 месяцев назад

Hi sir, I am new to this field, and I am running into an error AttributeError: module 'tensorflow.python.framework.graph_util' has no attribute 'convert_variables_to_constants' when running the build. Also, when I run the trainer_settings.yaml, I am getting a value error of input size. Any help is appreciated, and thanks for the video and knowledge. File "/Users/riloscode/anaconda3/envs/mlagents2/lib/python3.8/site-packages/tensorflow/python/client/session.py", line 1166, in _run raise ValueError( ValueError: Cannot feed value of shape (1024, 0) for Tensor optimizer/Placeholder_2:0, which has shape (?, 3)

@gabrielmaestrecosta628 Год назад

My computer has a compilation error, is it because I use archlinux?? I can't load the environment, it doesn't show anything on my screen when I load the project

@sanjithtron6000 Год назад

disable your anti virus

@thomasvaniseghem Год назад

I'd love to help but I don't have a lot of info and knowledge about Unity and Linux. I think it might have something to do with your Unity installation.

@irynatyshchenko6851 Год назад

Has something been changes? The car is unable to park

@adarshas8827 5 месяцев назад

I'm facing the same problem right now

@isolatedsushi5996 2 года назад

Awesome video! However what I don't understand is why each agent needs it's own parking lot for parallel parking. Why can't there be multiple cars all trying to park on the same spot if you disable their collisions. Could you explain this?

@thomasvaniseghem 2 года назад

That's actually quite a good idea! Don't know why I didn't think of that, it might be a good optimization. The only problem with this would be that all agents encounter the same traffic but personally I think that wouldn't be a huge problem. Might look into implementing this. Thanks!

@naroson9155 Год назад

Hey, I was wondering where to find the standard assets for the car controller scripts in the unity store? I want to get the latest version for a project Im working on

@mikhailhumphries Год назад

When you coming back?

@thomasvaniseghem Год назад

Graduated and started working a fulltime job so pretty bussy right now. I am however looking into some new interesting AI projects in that I might cover :)