A small 2D simulation in which cars learn to maneuver through a course by themselves, using a neural network and evolutionary algorithms.

Interested in how Neural Networks work? Have a look at my one-minute-explanation: automobilrevue.net/v/ckVEelVUM3ltdzQ

This simulation was implemented in Unity. You can find detailed information about how this simulation works, as well as a link to the entire source code on my website: arztsamuel.github.io/en/projects/unity/deepCars/deepCars.html

Don't miss any future videos, by subscribing to my channel.
Follow me on Twitter: twitter.com/SamuelArzt

Comments

    • Samuel Arzt
      Samuel Arzt

      In case you are interested in how Neural Networks work, I made a one-minute explainer: https://www.youtube.com/watch?v=rEDzUT3ymw4

      about 1 year ago
    • HR_
      HR_ In reply to Samuel Arzt

      How did you make the simulation?

      about 1 month ago
    • Trung Ng
      Trung Ng In reply to Samuel Arzt

      Is it really DL?

      about 1 month ago
    • Dark Knight
      Dark Knight In reply to Samuel Arzt

      What are inputs ?

      about 1 month ago
    • Christopher Hogben
      Christopher Hogben In reply to Samuel Arzt

      I've recently been reading up on RL and I was wondering if you can explain briefly how this is different from RL/DRL – or is it RL? Thanks in advance.

      about 2 months ago
    • D F
      D F In reply to Samuel Arzt

      ​+Jakob Mertens The vehicle is learning how to steer in order to avoid the edges of the road based on input from its sensors, and so it's not actually learning anything about the course itself (assuming no "over-training" has occurred, but that is another topic). Therefore, the vehicle can navigate any course based on what what it has learned from this single course. Hope that helps!

      about 2 months ago
    • KapSKY
      KapSKY

      Another episode of




      Youtube recomendations...

      about 1 day ago
    • laurent sergent
      laurent sergent

      ça me donne une idée ? cela peut servir pour la médecine ? implanté un émetteur à l’endroit d'une tumeur diffusant un signal à une certaine fréquence non nocive. pourrais appeler des nanobots qui pourraient se guider dans certains vaisseaux sanguins, d'un diamètre qui serait défini à l’avance jusque-là tumeur ; plus le nombre de nanobots serrait envoyé et venir se coller sur la paroi de la tumeur pourrait envoyer un signal du changement de la taille de la tumeur ?

      about 4 days ago
    • laurent sergent
      laurent sergent

      does that give me an idea? can this be used for medicine? implanted a transmitter at the location of a tumor diffusing a signal at a certain non-harmful frequency. could call nanobots that could be guided in certain blood vessels, of a diameter that would be defined in advance until then tumor; plus the number of nanobots would be sent and sticking to the wall of the tumor could send a signal of the change in the size of the tumor?

      about 4 days ago
    • rob miles
      rob miles

      seems like this can apply to human generations as well. Each generation learns something new. I may be stuck somewhere in the middle like the yellow car

      about 4 days ago
    • tig t
      tig t

      I cant understand the output. How can we express the sequential movement from 2 output??

      about 4 days ago
    • Qiang Qiang
      Qiang Qiang

      code ?

      about 5 days ago
    • Prateek 001
      Prateek 001

      Did u use genetic algorithm instead of backpropagation in order to tune the weights.. or did u use both?

      about 6 days ago
    • Samuel Arzt
      Samuel Arzt In reply to Prateek 001

      I used a modified genetic algorithm. There was no backpropagation involved.
      If you are interested in the code, there is a link to it in the video description.

      about 6 days ago
    • Prateek 001
      Prateek 001

      What are the desired outputs of the neural network in this case? Is it like...front sensor value should be highest and side ones be the lowest...?
      Please reply. I am also beginning to work on same kind of project

      about 6 days ago
    • Samuel Arzt
      Samuel Arzt In reply to Prateek 001

      +Prateek 001 In Unity anything that goes beyond very standard things is done with C# scripts. There are a lot of free official Unity tutorials available on their website. If you want to get into game-like simulations I'd recommend having a look at those tutorials.

      about 6 days ago
    • Prateek 001
      Prateek 001 In reply to Prateek 001

      +Samuel Arzt thanks for replying!one last query....how do I learn to simulate in unity....actually I am interested in autonomous stuff ....how do I learn to simulate environment and objects like u did in ur project?? U know like moving objects around and control them

      about 6 days ago
    • Samuel Arzt
      Samuel Arzt In reply to Prateek 001

      The outputs of the neural network are turning and engine force. The sensors are the inputs.
      The network solely learns by environment feedback (i.e. the evaluation function). Cars that progress further in the course are allowed to reproduce.
      In this sense, the desired output is the output which lead to a higher evaluation value.

      about 6 days ago
    • Mohannad مهند
      Mohannad مهند

      May I ask . Why you making this ? Deep learning for who and for what purpose?
      Thanks

      about 7 days ago
    • Samuel Arzt
      Samuel Arzt In reply to Mohannad مهند

      This particular project was for educational purposes. I wanted to learn and understand how neural networks work and how they can be trained.
      The end product was quite entertaining to watch, hence I uploaded a video of it. So in the end it also served entertainment purposes and also the purpose of getting people interested in the subject matter.

      about 7 days ago
    • meggi
      meggi

      i'd love to see a AI learn to play lemmings

      about 9 days ago
    • Fokosmok
      Fokosmok

      I'm not impressed

      about 10 days ago
    • 남윤성
      남윤성

      개멍청한거 같은데

      about 10 days ago
    • RedCore
      RedCore

      I'm third gender

      about 11 days ago
    • Arnaud de Broissia
      Arnaud de Broissia

      This is no Deep Learning, this is evolutionary algorithm. Two very different learning approaches.

      about 11 days ago
    • Pedro Lopevia
      Pedro Lopevia

      Where are the bets?

      about 12 days ago
    • Görkem Örgerim
      Görkem Örgerim

      Good video! I'm liked and share 875 times :D

      about 12 days ago
    • blazingax05
      blazingax05

      This reminds me of learning a new technique on the guitar

      about 14 days ago
    • Henry Cramer
      Henry Cramer

      Isn't this machine learning with an neural net, not deep learning, because the puzzle is stationary: "It gets better as it goes" which is the definition of machine learning.

      about 14 days ago
    • ReGaLLL C*
      ReGaLLL C*

      2:42 when hero shows up

      about 14 days ago
    • NBG. -PhAntOm
      NBG. -PhAntOm

      I suggest implementing 4 output neurons, for extra precision ( left,right, middleleft, middleright).

      about 15 days ago
    • χρηστος Μπακολουκας
      χρηστος Μπακολουκας

      GOZMOTE TSAKNIA

      about 15 days ago
    • José Russo Cassignol
      José Russo Cassignol

      Thanks for the info in the description!

      about 16 days ago
    • Grinder Arts
      Grinder Arts

      Am i the only one to find this disturbing?

      about 16 days ago
    • BRGD EnvY8
      BRGD EnvY8

      Are we using binary trees in here?

      about 16 days ago
    • Samuel Arzt
      Samuel Arzt In reply to BRGD EnvY8

      This relies only on a 4 layered feedforward neural network and a modified genetic algorithm to train it.

      about 15 days ago
    • Sungwon Shin
      Sungwon Shin

      red car looks like my life in a nutshell

      about 17 days ago
    • Etzyy
      Etzyy

      Tesla be like:

      about 17 days ago
    • Mondschein91
      Mondschein91

      Damn Skynet

      about 17 days ago
    • HYDRV
      HYDRV

      Nice AI

      about 17 days ago
    • Lang Kuai
      Lang Kuai

      Deep learning YouTube videos.
      Conspiracy alarm

      about 17 days ago