Hey guys! Wanted to share my new devlog on my Unity game where I am training an AI to learn to dodge bullets using ML-Agents/Reinforcement Learning! I am kinda amazed how good it plays after proper training…
I had been having issues with training on a high time-scale and testing on time-scale 1. For “reasons”, testing on lower time scales was always doing much much worse. I found a little workaround where I replaced all my coroutines with FixedDeltaTime-dependent delays. Full explanation and game progress here:
