This section holds results about the agents: training, perception and how their strength is measured. A result gets a page once it stands on its own: the question, the method, the result with how sure the project is, and what it doesn't settle. Pages are written once, and amended only when a later result supersedes or refutes them.
Note
No match has been rated yet. Two hand-written AI players now play full matches on some of the project's arenas, through the same interface every agent will use, recorded. See measurement and the roadmap. The first experiments below, learning to move, ran in a toy simulation off the server and claim nothing about playing strength.
Results#
| Page | What it covers |
|---|---|
| Learning to move | The first training experiments: small learning programs taught to walk and chase in a simple simulation on one computer, each with its prediction written down first. |