Customizing Your Own Task

To create a custom task, follow these steps:

• Set up a new task environment.
• Generate a configuration file for each task variant.

Creating a New Task Environment

1. Clone the Task Template
   Copy the task_build folder located under teamcraft/teamcraft/tasks/, renaming it in the format task_xxx.

2. Rename the Minecraft World Save Folder
   Inside task_xxx, rename the Minecraft world save folder (world_build) to world_xxx.

3. (Optional) Modify the Minecraft World

   • Edit world_xxx using a local Minecraft instance (refer to env doc under Visualization for setup instructions).
   • Load world_xxx as a game save.
   • Modifed the world using creative mode (Use command /gamemode creative)
   • Save the world

4. Modify the build_env.py File

   • Rename the BuildEnv class following the format xxxEnv.
   • Adjust any internal parameters using env doc as a reference.
   • Modify the calculate_reward function:
     • For block-based rewards, refer to BreakEnv and BuildEnv.
     • For inventory-based rewards, refer to FarmEnv and SmeltEnv.
   • Update the self.done condition.

5. Update Import Statements
   Modify the following __init__.py files to import your environment class:

   • teamcraft/teamcraft/tasks/task_xxx/__init__.py
   • teamcraft/teamcraft/tasks/__init__.py
   • teamcraft/teamcraft/__init__.py

Generating the Configuration File

1. Follow the step by step instructions in Customize Your Task.ipynb
   • Alternatively, you can also follow a complete python file, config_gen.py, sarting from line 170, to generate the configuration file.
2. Replace the existing .json file under task_xxx/configure with the new configuration.

Run

Once completed, you can import and use your custom class with:

from teamcraft import xxxEnv

env = xxxEnv() 

total_steps = 5
total_variants = 250

for seed in range(total_variants):
    reset_info = env.reset(seed)
    task_images = reset_info[0]
    images, state, inventory, done, reward = reset_info[1]

    for _ in range(total_steps):
        actions = "your action here"
        images, state, inventory, done, reward = env.step(actions)

env.close()