I used to play Toribash a long time ago and had forgotten all about it. More recently I have been working on neural networks and the idea hit me to try to evolve a neural network to play toribash. The outputs for the network are pretty clear. 20 two bit outputs to control the joints then two single bit outputs for the states of the hands. What remains unclear is what inputs I should use. I was hoping a bot maker could help fill me in on what the bots typically take into consideration when making their moves. Additionally I could make the AI evolve much quicker if I could speed up the animation time or even run it without the animations at all. Would that be possible with toribash or will I have to settle for the normal playback speed?

Moved this to scripts since this clearly doesn't concern "mods" as we call them around here

~ Good luck on your quest

If you managed to make the AI auto-restart the replay it's in, then you can use "<" or ">" to adjust playback speed. My only concern is that if the bot is alone with no goals, what would it do? Normally you give the AI a simple task and it will teach itself to accomplish it within your parameters. Google does this with their AI found here.
Also, if you make him fight another character, the joint positions in Toribash's default ".rpl" files would become conflicting. If you dive further into Toribash's code I'm sure you could do a lot of re-writing and accomplish your goal. If this works then you'll be a legend in the community. Good Luck!