The goal of our project is to segment and label different structures and items in minecraft as our player walks through an environment. An example of a structure that would be classified would be a tree or a sky with clouds. Every important detail in the player’s view should be labeled correctly and efficiently. If time permits, I will attempt to include minecraft battle royal using reinforcement learning.
Our plan is to generate a world and then scan through the world and label structures and objects.
Using Unet for semantic segmentation, deep Q learning for reinforcement learning.
Start by scanning the entire screen from the top left and returning the block type from each pixel
The success of the project can be determined in a similar manner to our performance validation where classification and location accuracy success can be evaluated. The project can also include a runtime output that will measure the efficiency. The runtime should decrease as the player walks through the minecraft world as part of a learning algorithm.
10:00am - 10:15am, Thursday, October 22, 2020
Using Malmo functions, figure a way to detect the type and location of a block that appears on the screen.