vladfi1

It was just basic stuff like player positions, velocities, and animation states.

In starcraft there's a much bigger advantage since humans are inherently "single threaded", and so you can get much bigger discrepancies in APM (or EPM). Smash is more like 1 unit vs 1 unit micro. The precision and timing are still advantages for the AI, but not so much raw parallel compute.

Basically we tried that and it sort of works, but performance degrades pretty fast with each frame of delay. The issue is likely that it makes credit assignment much harder. Instead of seeing an immediate change in the state (which your critic can interpret as good or bad), you have to wait a bunch of frames during which your previous actions are taking effect and interfering with the reward signal.

The bots are given a very reduced action set. The Falcon master bot actually had only cardinal directions on the control stick - later bots were also given the 4 diagonals.

It does not see the opponent's actions before they take effect on screen, and the actual controller states are not part of the feature representation we used (though they actually are somewhere in the RAM).

The main blocker on using pixels has been getting them from the emulator. I doubt pixels would give a big advantage over RAM features, especially after projectile and stage info is added (there's a PR pending). Captain Falcon (the main character used) doesn't have projectiles anyways.

In fact, RAM features are likely to be much more useful for model-based approaches, which may be important for solving the action-delay problems.

As for multiple characters, the character ID is available to the network. I doubt pixels will be help there either.