Upvote Upvoted 14 Downvote Downvoted
really cool video about tf2 reinforcement learning
1
#1
0 Frags +

https://www.youtube.com/watch?v=3kf_Hu14LOQ

wasnt there a thread recently where something similar was talked about?

https://www.youtube.com/watch?v=3kf_Hu14LOQ

wasnt there a thread recently where something similar was talked about?
2
#2
Fireside Casts
14 Frags +

didnt know sin k4rma had a brother

didnt know sin k4rma had a brother
3
#3
15 Frags +

https://i.imgur.com/n9aHVzS.png

[img]https://i.imgur.com/n9aHVzS.png[/img]
4
#4
0 Frags +

I wonder if one day we get TF2 AI to simulate 6s and fight out a few dozen strategies and tactics; to objectively show which is the best counter for which

I wonder if one day we get TF2 AI to simulate 6s and fight out a few dozen strategies and tactics; to objectively show which is the best counter for which
5
#5
0 Frags +
RahmedI wonder if one day we get TF2 AI to simulate 6s and fight out a few dozen strategies and tactics; to objectively show which is the best counter for which

might be interesting, but if you get to the level where they are much better than humans it's probably not going to be useful for normal play. Ex: Computers can get 100% acc meatshots which is an unrealistic assumption for human players.

[quote=Rahmed]I wonder if one day we get TF2 AI to simulate 6s and fight out a few dozen strategies and tactics; to objectively show which is the best counter for which[/quote]

might be interesting, but if you get to the level where they are much better than humans it's probably not going to be useful for normal play. Ex: Computers can get 100% acc meatshots which is an unrealistic assumption for human players.
6
#6
0 Frags +
plumRahmedI wonder if one day we get TF2 AI to simulate 6s and fight out a few dozen strategies and tactics; to objectively show which is the best counter for which
might be interesting, but if you get to the level where they are much better than humans it's probably not going to be useful for normal play. Ex: Computers can get 100% acc meatshots which is an unrealistic assumption for human players.

You could set it at 70% and have a bias of increasing/decreasing in certain scenarios

[quote=plum][quote=Rahmed]I wonder if one day we get TF2 AI to simulate 6s and fight out a few dozen strategies and tactics; to objectively show which is the best counter for which[/quote]

might be interesting, but if you get to the level where they are much better than humans it's probably not going to be useful for normal play. Ex: Computers can get 100% acc meatshots which is an unrealistic assumption for human players.[/quote]

You could set it at 70% and have a bias of increasing/decreasing in certain scenarios
7
#7
2 Frags +

state space is so much bigger in 6s tf2 compared to other RL applications i see in games that i'm pretty sure tf2 will be completely dead before we reach that point lol

state space is so much bigger in 6s tf2 compared to other RL applications i see in games that i'm pretty sure tf2 will be completely dead before we reach that point lol
8
#8
2 Frags +
plumEx: Computers can get 100% acc meatshots which is an unrealistic assumption for human players.

if elijah can do it, so can you

[quote=plum]Ex: Computers can get 100% acc meatshots which is an unrealistic assumption for human players.[/quote]
if elijah can do it, so can you
9
#9
0 Frags +
springrollsstate space is so much bigger in 6s tf2 compared to other RL applications i see in games that i'm pretty sure tf2 will be completely dead before we reach that point lol

openai did RL for dota2, but you're right, it took millions of dollars to train

[quote=springrolls]state space is so much bigger in 6s tf2 compared to other RL applications i see in games that i'm pretty sure tf2 will be completely dead before we reach that point lol[/quote]
openai did RL for dota2, but you're right, it took millions of dollars to train
Please sign in through STEAM to post a comment.