We want to use AI/ML to create a farm in the world. Our idea is to make it easier for players to grow food for consumption or feeding. Given an inventory of seeds for wheat, carrots, potatoes, a gardening hoe, and buckets of water, the agent will figure out how to grow crops. It looks at its grid in order to see the world state, which will show which land can grow food, and which foods are ready for harvest. The application for this project is giving the player a stable food source.

Reinforcement learning with Markov Models

The metrics are how much food you can get by the end of 5 days in the inventory in a 20x20 block enclosure. The baseline will the the number of crops in its starting inventory. We expect out inventory to at least double in size based on the growth rates of crops and probabilities of crop yield. We will visualize the internals of the algorithm with a QTable that will have different colors based on what it learns is good or bad. For example, farming requires water, but if it falls into the water, it will be bad, so the actions around them that lead to the water will be red. If it harvests a crop early, it would be yellow because it’s getting more seeds. If the agent harvests the crop when the crop is fully grown, it will be green. In the moonshot case, the agent will understand that plants require water and light and will utilize their block to the fullest and find the best orientation of water pits and torches to maximize the speed and quantity of growing; the agent will also be able to recognize the best time to harvest the crops in order to maximize the highest crop yield and reward.

Tuesday, April 25th. Time - 1:15 pm