"Creatures" is a stretch given that the environment is Minecraft, but the idea is simple: iteratively add blocks conditionally (tensor convolution) on the current environment (blocks) to maximize some reward. In this case I use PPO RL to train creatures to touch a glowstone block but you can adapt it to use any algorithm and reward (easily, as it uses the Ray framework).

What I like about this work: iteratively finding solutions has a long and colorful history of doing things well: gradient boosting, ResNets, Stable Diffusion, etc. We're after some end optimal state and usually try to get right to it via some flavor of SGD. Baking in improvements as part of the problem structure is helpful. I like to think of this as solving the derivative of the solution with respect to time and then integrating over time. The Minecraft environment naturally prevents overfitting because you can spawn creatures in a diverse array of environments and train one policy over them all.

What I don't like about this work: Minecraft was chosen for computational reasons but doesn't make cool creatures that walk around and look like cheetahs or whatever. I've compared it against nothing, but it works real good! <g>. Videos in the README.

You can install / run with Docker and observe metrics with Tensorboard. Should be plug and play. If not comment. Enjoy!