MC-Planner DEPS Simulator

This project Simulates DEPS

Project Structure

.
├── src/
│   ├── minedojo_core/          # Core minedojo components (data, sim, tasks)
│   ├── gymnasium_env/          # Gymnasium environment wrapper
│   ├── models/                 # Model implementations
│   ├── utils/                  # Utility functions
│   └── __init__.py
├── configs/                    # Configuration files
├── data/                       # Data files and prompts
├── main.py                     # Main execution script
├── planner.py                  # Planning module
├── selector.py                 # Selection module
├── controller.py               # Controller module
├── requirements.txt            # Python dependencies
└── README.md

Features

• Java Dependency Removal: Minimized Java dependencies outside Minecraft MDK
• Gymnasium Integration: Migrated from MineDojo to gymnasium environment
• Modular Design: Separated core components for better maintainability
• Configuration-based: Experiment configuration through YAML files
• Optimized Execution: Streamlined for research and experimentation

Installation

# Install dependencies
pip install -r requirements.txt

# Install in development mode
pip install -e .

Usage

# Run with default configuration
python main.py

# Run single task
python main.py eval.single_task=true eval.task_name=obtain_wooden_slab

# Run with custom configuration
python main.py --config-path configs --config-name custom

# Test core functionality
python -c "
import sys; sys.path.append('src')
from src.gymnasium_env import MineDojoEnv
from planner import Planner
from selector import Selector
print('All modules working correctly')
"

Quick Start

1. Clone and install:

   git clone <repository>
   cd Simulator-master
   pip install -r requirements.txt

2. Test the installation:

   python -c "from src.gymnasium_env import MineDojoEnv; print('✓ Installation successful')"

3. Run a simple task:

   python main.py eval.single_task=true eval.task_name=obtain_wooden_slab

Local LLM Setup

For planning functionality, you can use a local LLM:

# Set environment variables
export LLM_API_BASE="http://localhost:8000/v1"
export LLM_MODEL="local-llama3"
export LLM_API_KEY="DUMMY"

# Or edit data/openai_keys.txt

Components

Core Modules

• main.py: Main entry point and experiment orchestration
• planner.py: LLM-based planning with local model support
• selector.py: Goal selection and horizon planning
• controller.py: Action execution and environment interaction

Configuration

All experiments can be configured through YAML files in the configs/ directory.

Data

The data/ directory contains:

• Task definitions and prompts
• Goal mappings and libraries
• Pre-computed embeddings

Development

This project is designed for research in multi-task agents using large language models in Minecraft environments.