AI Learns To Parkour
HTML-код
- Опубликовано: 8 ноя 2024
- In this video Marshy attempts to parkour his way across various obstacles with the use of deep reinforcement learning.
This is my first time using unity so forgive me if it looks janky. I made use of unity's ml-agent package to train the ai with deep reinforcement learning. It's brain is made up of 4 layers of 128 hidden units each. It takes in various observations such as its position, rotation, the 'goal', and many raycast projections to mimic vision. Any feedback would be greatly appreciated, thank you!
Subscriber count: 66
The budget increased let's go!!!
Haha 😁
amzing video!
Thank you!! 😁
great video again :D
Thank you! 🙏🏻
What input change is it waiting for in Level 3? surely if its standing still it would continue to output the stand still command?
(unless there's a timer input or the raycasts are semi random? maybe the outputs are only allowed to change every so often and its waiting for the next output cycle??)
One thing I did was set the ai to only request a decision every 20 academy steps. However I think where the ‘waiting’ mostly came from was lag - sometimes the programme would freeze for a second or two when I was recording it because I had many simultaneous environments in the background. However I changed this in level 4 onwards so it would lag less. Hope this helps! :D
This is cool
Thanks!
I really enjoyed video. Could you please share the source code ? It would be really helpful to see how you implemented it. Also, if possible, could you provide a brief demonstration of how to run the code? Keep up the good work ,thanks a lot!"
Hi, glad to hear that you enjoyed the video! The source code is so messy so I don’t think I’ll show it 😅 However if you’re interested, perhaps some time in the future I’ll make a tutorial on how to get started with ml-agents for ai reinforcement learning 😄
You look so cute