Very intuitive and a great tool to introduce someone to reinforcement learning. Excited to play the full version of this when released!
Feedback:
- Show volume settings during tutorial - I had to quickly skip the tutorial because the narrator was too loud and there wasn't a way to adjust volume in-game.
- In training mode the bot will sometimes 'spit' out a ball into the air - I wasn't sure what this is supposed to indicate.
- Include a way to reset a bot's brain to forget all prior training. Also, as a suggestion, maybe the best performing brain (maximum 'love' / minimum 'fear') could be saved so you can revert to it. Right now the last 10 brains are saved but it'd be nice if a brain could be protected from being overwritten - something I've seen in other reinforcement learning games like boxcar2d.
Bugs:
- The 'reset world' button stopped working after the first session - after re-opening the game, clicking it doesn't do anything and the bots that died don't respawn. This persists despite restarting the game.
- The 'raise' and 'lower' tools in training mode can be brought into the overworld by having them selected in training mode and then clicking 'return to overworld'.
- I have repeatedly encountered a bug where when I click on the overworld it teleports a bot/item to where my cursor is. I believe this issue is related to #2 (maybe I had a bot/item selected in training and it carried over) but I can't reproduce it consistently.