ML agents, is for machine learning using python to create artificial intelligence.
first ml agents can move, jump , find objects on a large map by himself using neuronal network algorithm.
median at 0.920. std 0.120 with around 4820000 training.
very fascinating to work with machine learning
first was moving , jumping and making tasks.
second game was first getting the objective, running until the goal. median 39.978 100% is 39 ???:)
than it was to move around obstacles with 39 as 100% but with obstacles that takes down points, more difficulties..... still in process.
after over 15 errors, bad test, i finally understand the methodology: be super analytical of all variables and details :collider, rewards, (lower when touching obstacles -0.2f....
mean of26 , 27 with random . set up is very important , and you can influenced the ai to get the result you want . if the first 100000 steps are not good median improvment, you can restart the set up lol .
hide and seek game with AI
getting really good value loss quickly, "happy", getting good at it.
with 2 no symmetrical machine learning teams, each team compete against each other.
-with increase difficulty over time ( explained big changes in policy loss ) but they get smarter and smarter :
-more hideout, hiding places at random places etc...
cool team strategies behavior observed :
1. the red team try to surround the blue team with lasers (raycast) and finish them .
2. the red team position their players in different spots to cover the place.
3. the red team (seekers) hide under walls to be not detectable and strike. lol fun
blue team (hide)
1. one player cover the vision( raycast) and others on the back.
2. blue get on tha back of the red payers to don't be seen.
.3.tendency top stay in group to raycast (vision) at 360 degrees.
4. some agents stay on the border of the map to be undetected lol , really.
the difficulty when working with 2 teams, they both get better to evitate each other, so it 's more difficult to find improvments .
