# MineRL topics

Minecraft that might be of interest:

~~1. 自動化農業和資源管理~~
~~- 自己做機關，類似小麥收割機~~
## 挑主題
- ==決定：[跑酷](https://hackmd.io/mhCHg0eISJ-0610ultRwYQ)==
1.可實際在mineRL操作
2.AI學習相關
(3.) 可延伸至現實
1. （森）自適應跑酷和越障導航(中)
- [AI Learns to Walk deep reinforcement learning](https://www.youtube.com/watch?v=2tamH76Tjvw)
- 難點：自訂地圖
2. (沛)打一群怪，武器是弓箭
- 打移動物，打中給reward
- 應用在現實：PID控制 輔助or替代
- 難點：需要自訂地圖
~~5. 自主尋寶探索~~
~~- 搜集材料獲得不同reward~~
3. 學習與Minecraft 村民互動和管理以及交易
- 設計獎勵函數來引導代理學習。激勵代理:
- 最大化交易收益。
- 優化資源收集和交易。
- 通過提高村民經驗解鎖更好的交易。
- 確保村民福祉以維護繁榮的村莊
- 難點：可以生成在村莊旁邊嗎，夜晚被殭屍襲擊，蒐集材料很困難
4. （許）物品分類(ex: 給一定的)
- 應用在現實:物品分類
- 難點：
5. (翔)給定一個範圍(ex: 16 * 16)，把agent看到的視野變成一張圖片，用對應顏色羊毛畫出(難)
- 應用在現實：圖型辨識、變形視角校正
- chatGPT有生成一些code，應該是可以透過mineRL實作
-

[Simple Reinforcement Learning example in Minecraft](https://tsmatz.wordpress.com/2020/07/09/minerl-and-malmo-reinforcement-learning-in-minecraft/)
### Sources
https://minerl.readthedocs.io/en/latest/
https://stackoverflow.com/questions/38034754/error-installing-getch-module-python-3-5
https://www.youtube.com/watch?v=G44Lkj7XDsA&ab_channel=Hung-yiLee
## Idea_1: Reinforcement Learning for Efficient Exploration and Navigation in Minecraft

Objective: Develop a reinforcement learning agent that learns to efficiently explore and navigate complex environments in Minecraft.

Overview:

1. Set up a Minecraft environment for your agent to interact with. You can use the MineRL library (https://minerl.io/), which provides a Python interface to Minecraft for reinforcement learning research.

2. Create or select diverse and challenging environments for your agent to explore and navigate. These can include environments with varying terrain types, obstacles, and hazards, such as forests, mountains, caves, and underwater areas.

3. Develop a reward function that will guide your agent's learning process. The reward function should encourage the agent to explore new areas and reach specific target locations as quickly and safely as possible. It can include factors such as:

- Positive rewards for visiting unexplored areas or reaching target locations
- Negative rewards for falling into hazards (e.g., lava, water) or taking too much time to explore/navigate
- Bonuses for discovering rare resources or structures

4. Implement a reinforcement learning algorithm for your agent. You can choose from existing algorithms like Q-learning, Deep Q-Networks (DQN), or Proximal Policy Optimization (PPO), depending on your familiarity with reinforcement learning.

5. Train your agent in the Minecraft environment using the reinforcement learning algorithm and reward function. The agent should learn to efficiently explore the environment, avoid hazards, and navigate to target locations.

6. Evaluate your agent's performance by measuring how quickly and effectively it can explore new environments and reach target locations. You can compare the agent's performance to human players or other reinforcement learning agents.

7. (Optional) Extend your project by adding more complexity, such as:

- Introducing multi-agent exploration, where multiple agents collaborate or compete to explore the environment
- Allowing the agent to learn to use tools or equipment (e.g., boats, elytra wings) to enhance its exploration and navigation capabilities
- Implementing a hierarchical reinforcement learning approach, where the agent first learns low-level navigation skills and then combines them to achieve high-level exploration goals

This project will provide you with hands-on experience in reinforcement learning and allow you to tackle the interesting challenges of exploration and navigation in Minecraft's diverse and complex environments.

## Idea_2: Reinforcement Learning for Automated Resource Gathering and Crafting in Minecraft

Objective: Develop a reinforcement learning agent that learns to efficiently gather resources and craft items in Minecraft.

Overview:

1. Set up a Minecraft environment for your agent to interact with. You can use the MineRL library (https://minerl.io/), which provides a Python interface to Minecraft for reinforcement learning research.

2. Define a set of items or resources that the agent should be able to gather and craft. These can range from basic resources like wood, stone, and iron to more complex items like tools, weapons, and armor.

3. Develop a reward function that will guide your agent's learning process. The reward function should encourage the agent to gather the necessary resources and craft the target items as quickly and efficiently as possible. It can include factors such as:

- Positive rewards for successfully gathering a resource or crafting an item
- Negative rewards for inefficiencies, like wasting resources or taking too long to gather/craft
- Time-based penalties to encourage the agent to complete tasks faster

4. Implement a reinforcement learning algorithm for your agent. You can choose from existing algorithms like Q-learning, Deep Q-Networks (DQN), or Proximal Policy Optimization (PPO), depending on your familiarity with reinforcement learning.

5. Train your agent in the Minecraft environment using the reinforcement learning algorithm and reward function. The agent should learn to explore the environment, gather the necessary resources, and craft the target items.

6. Evaluate your agent's performance by measuring how long it takes for the agent to gather resources and craft the specified items. You can also compare the agent's efficiency to human players or other reinforcement learning agents.

7. (Optional) Extend your project by adding more complexity, such as:

- Allowing the agent to learn to gather and craft a wider variety of items
- Introducing a multi-agent setup, where multiple agents collaborate or compete to gather resources and craft items
- Adding an element of exploration, where the agent must learn to navigate and find resources in procedurally generated environments

This project will provide you with hands-on experience in reinforcement learning and allow you to explore the interesting challenges of resource gathering and crafting in Minecraft.