Reinforcement Learning and Minecraft | Lars Gregori | ML Con 2018 Spring

Learn how reinforcement learning can be applied to the popular game Minecraft, using a combination of image recognition and Q-learning algorithms to train an agent to navigate and find a blue block.

Key takeaways
  • Reinforcement learning is a type of machine learning where an agent learns to make decisions by interacting with an environment and receiving rewards or penalties.
  • Minecraft can be used as a platform for reinforcement learning, allowing agents to learn to navigate and interact with the game world.
  • The speaker used a combination of image recognition and reinforcement learning to train an agent to navigate a Minecraft world and find a blue block.
  • The agent was trained using a Q-learning algorithm, which learns to predict the expected return or reward for each action taken in the environment.
  • The speaker used a discount rate (gamma) to determine how much to discount future rewards, and an alpha value to determine how much to update the Q-values based on new experiences.
  • The agent was able to learn to navigate the Minecraft world and find the blue block, but struggled with scenarios where it had to make decisions about which path to take.
  • The speaker used a combination of trial and error and reinforcement learning to train the agent, and was able to achieve good results with a relatively simple algorithm.
  • The speaker also discussed the use of deep reinforcement learning, which uses neural networks to learn the Q-values and can be used to solve more complex problems.
  • The speaker used the Project Malmo platform, which is an open-source platform for building and testing reinforcement learning agents, to train the agent.
  • The speaker also mentioned the use of other machine learning techniques, such as supervised learning and unsupervised learning, and discussed the differences between these approaches.
  • The speaker used Python as the programming language for the project, and mentioned the use of other languages such as Lua and Java.
  • The speaker also discussed the use of reinforcement learning in other domains, such as IoT data visualization and microservices for shopping platforms.