Notifications
bg
David Busch
Developer, Designer - Programmer
Austin, United States
37
Likes
19
Followers
8
Connections
All
Articles1
Games1
Showcases0
Column2
Jobs0
David Busch
updated a submission to ML-Agents Challenge I
Jan 30, 2018 11:48 PM
Hide / Escape - Avoidance of Pursuing Enemies
Training ML Agents to Avoid Traditional AI Using Curriculum-Based Reinforcement Learning
Article
View Challenge
View Challenge
David Busch
10 months ago
Dahyun KimAh, I see. Thank you so much! I actually thought that the rewards should always be between -1 and 1, now I see that wasn't true.
I think the rule of thumb is similar to the state inputs. They are safest being from -1 to 1, but with my experience, you can probably go up to -99 to 99 before the normalization really becomes a factor
DK
Dahyun Kim
10 months ago
Ah, I see. Thank you so much! I actually thought that the rewards should always be between -1 and 1, now I see that wasn't true.
David Busch
10 months ago
Dahyun KimNice work. Thanks for the precise explanation. It's very helpful. I'm also working on the Unity ML-agents, trying to make a little game. I'm wondering if I can take a look at your Unity Script of the Agent and the Academy, if possible. The little game I'm making is very similar to yours. And in your video, does 'scores' mean 'mean rewards'?
Yes, scores and rewards are the same in this case. Of course you can look at my scripts! My entire project is on github and is listed on the project page. I'll post it here as well: https://github.com/HappySlice/hide-escape
DK
Dahyun Kim
10 months ago
Nice work. Thanks for the precise explanation. It's very helpful. I'm also working on the Unity ML-agents, trying to make a little game. I'm wondering if I can take a look at your Unity Script of the Agent and the Academy, if possible. The little game I'm making is very similar to yours. And in your video, does 'scores' mean 'mean rewards'?
David Busch
uploaded a submission to ML-Agents Challenge I
Jan 29, 2018 11:21 PM
Hide / Escape - Avoidance of Pursuing Enemies
Training ML Agents to Avoid Traditional AI Using Curriculum-Based Reinforcement Learning
Article
View Challenge
View Challenge
David Busch
10 months ago
Dahyun KimAh, I see. Thank you so much! I actually thought that the rewards should always be between -1 and 1, now I see that wasn't true.
I think the rule of thumb is similar to the state inputs. They are safest being from -1 to 1, but with my experience, you can probably go up to -99 to 99 before the normalization really becomes a factor
DK
Dahyun Kim
10 months ago
Ah, I see. Thank you so much! I actually thought that the rewards should always be between -1 and 1, now I see that wasn't true.
David Busch
10 months ago
Dahyun KimNice work. Thanks for the precise explanation. It's very helpful. I'm also working on the Unity ML-agents, trying to make a little game. I'm wondering if I can take a look at your Unity Script of the Agent and the Academy, if possible. The little game I'm making is very similar to yours. And in your video, does 'scores' mean 'mean rewards'?
Yes, scores and rewards are the same in this case. Of course you can look at my scripts! My entire project is on github and is listed on the project page. I'll post it here as well: https://github.com/HappySlice/hide-escape
DK
Dahyun Kim
10 months ago
Nice work. Thanks for the precise explanation. It's very helpful. I'm also working on the Unity ML-agents, trying to make a little game. I'm wondering if I can take a look at your Unity Script of the Agent and the Academy, if possible. The little game I'm making is very similar to yours. And in your video, does 'scores' mean 'mean rewards'?
David Busch
published the game
Sep 29, 2017 7:48 AM
Drive Through Explosions
Virtual Reality Post-Apocalyptic Vehicular Combat
Game
About Me
I have been developing with Unity since 2014 with the Oculus DK2.
Booz Allen Hamilton
Immersive Engineer
Arizona St
Electrical Engineering
See more
Certifications (1)
Import
Unity Certified Developer
Sep 2017
See all
David Busch's liked projects (2)
Following (11)
Followers (19)
Following Companies (0)
Not following anyone yet
Groups (0)
No groups