Show this thread
A fun way to learn about neural networks and AI is to implement a simulation game giving your agents little neural net brains, and training them using a simple method like evolution. This demo trains a small neural network to drive around the track after only a few generations:pic.twitter.com/k5yyHJFAhu
-
-
-
If you are interested conducting this type of experiment, there's a similar environment that comes with
@OpenAI Gym. You'll need to hack it if you want the inputs to be simple LIDAR inputs, like the demo by the author of this video, rather than raw pixels.https://gym.openai.com/envs/CarRacing-v0/ …Show this thread
End of conversation
New conversation -
-
-
Fun fact: the neural network is not even necessary, a simple linear model and a genetic algorithm already work :) I've been doing something similar on real (Arduino-controlled) wheeled robots, it's super cool!
-
I’ve also got a linear controller to work on a car racing task (from pixel inputs) too recently :)https://twitter.com/hardmaru/status/916849748337876992?s=21 …
End of conversation
New conversation -
-
-
Here's a related demo that works on the web browser:https://twitter.com/hardmaru/status/796767214003224576 …
End of conversation
-
-
-
Fantastic. Are you going to post the source? I'd like to chuck some of the
#stablebaselines at it and maybe rank them. -
I didn't make this demo. See the original youtube video in the thread if you want to reach out to the author - the name is at the end of the video.
-
Next time please put credits directly in your tweet. Do not let your viewers believe that you are the author. That´s unfair.
End of conversation
New conversation -
-
-
mentioning neural network, but describing a genetic algorithm?
-
There is no contradiction here. The genetic algorithm can be used to search through the possibility space and find the optimal weights or any other parameter of the network.
End of conversation
New conversation -
-
-
And I still struggle to get two banks to agree on an schema for exchanging data.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
How does the trained system cope when you give it a different, unfamiliar track to drive around? Is the skill it has built transferable? Can you teach it to drive on any arbitrary track the first time it encounters it?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
here is my same implementation: https://www.youtube.com/watch?v=VSEAuLm3AQo … the code:https://github.com/kargarisaac/Reinforcement-learning/tree/master/RL_car …
-
How can such a system deal with the walls moving slowly? or random obstacles appearing on the road? Would it eventually handle them? Or is it only suitable with a reasonably stable environment?
-
It uses the sensors and can consider a penalty for the measurements and the distance to the walls.
End of conversation
New conversation -
-
-
Hah, this reminds me of a fun project from too long ago: NeuroRacer Video: https://youtu.be/30x1-bcpmNY PDF: http://ml.informatik.uni-freiburg.de/_media/publications/kr09.pdf …
-
Kietzmann, T.C., & Riedmiller, M. (2009). The Neuro Slot Car Racer: Reinforcement Learning in a Real World Setting, ICMLA 2009, Miami Beach, USA
End of conversation
New conversation -
-
-
Nice, this is very similar to how we did car AI in the Real Racing series. Brings back fond memories. Credit to Kynan Woodman who did most of that work AFAIK.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Too bad we can’t do this with government policy.
-
There is no "we", so of course "we" can't, but the ruling class can, and does, and it's working just great for them. The mistake is to believe the propaganda that the system is designed for "us" and that any issues mean it is somehow "broken" and ever more system is needed.
- 1 more reply
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.