I’ve always wanted to make an AI that learns how to juggle without any explicit instruction. I gave it a shot over the last couple of days and I’m going to have to shelve this for now to work on other stuff. This project has been a longstanding goal of mine, and I’m definitely going to come back to this and finish it at some point.
It’s made in Unity and is implemented using reinforcement learning, similar to our fish project. There’s only one ball because I at least wanted to get that passing correctly back and forth before moving on.
I definitely could have gotten it to work better than that, but part of the goal is to have an AI learn to juggle without instruction. I could start adding “hints” like telling it where it should aim the ball or how hard the ball should be it, but every time I do that is adding in a little bit of cheating to the core idea of the project. I’d rather call it done for now and come back to it later than end up just manually making the thing effectively juggle.
Anyway, code is here if you want to look at it. If I ever get it working, I’ll post it here. It was going to be a Unity WebGL project, so it will run in your browser and slowly learn to juggle over time.