AI and Minecraft
summary
Uses Project Malmo and Minecraft as an experimental RL environment, controlling an agent via XML and Python and training with Q-learning and a small CNN in Keras. Builds a grid cliff-walk scenario (100 random runs, 12 action classes) and shows alpha/gamma effects on convergence; agent learns a clear path after dozens of episodes but sometimes gets stuck in local minima or falls into lava. Implements image preprocessing by subtracting two frames so a CNN with two-frame subtraction can predict moves; results are promising but noisy.
AI-written summary · about
