ML Agents train at runtime?

Can agents be trained at runtime ? or do any form of learning at runtime once trained?

I ask as I would like to add machine learning into our game but need the agents to learn from being killed by the player, hence need some sort of runtime feedback

3 Likes

I am very interested in this topic too

as far as I know. No, not yet. But you can still collect data from your players (you need to create your own data collection implementation), and train your agents yourself and send new trained data to your players once a while. like daily or weekly intervals.

to confirm what NorthStar79 said, no not yet, I go into a bit more detail in this thread here . you can use your created AI in exported projects though, and use your inputs as a developer with python installed to manually train your AI with heuristics! the documentation explains a bit more about imitation learning.

I’ve been thinking about this and stumble on this thread too.

I think it’s possible if you’re willing to keep a server instance running where you send new data over and then use those data to train it. But this means, for each profile, you’ll have to keep track of their progress.

1 Like