I trained ML agents to play Tag . The results are so unexpected | Official Unity Discord | Page 1

YouTube

All things GAME!

I trained two AI agents in a simple arena.

No scripted behavior.
No hard-coded rules.
Just reinforcement learning.

At first, they didn’t understand walls, space, or each other.
They failed. A lot.

Then I introduced a single object — the Crown.

What happened next wasn’t something I explicitly trained for.

One agent stopped trying to wi...

▶ Play video

#I trained ML agents to play Tag . The results are so unexpected