KAIROS MineRL BASALT

Codebase for the solution that won first place and was awarded the most human-like agent in the 2021 NeurIPS Competition MineRL BASALT Challenge.

Original README of the competition: https://github.com/minerllabs/basalt_competition_baseline_submissions

Installation Guide

Follow the steps below to install Anaconda, create Anaconda environment (conda-env create -f environment.yml conda activate basalt), then activate the conda environment (conda activate basalt), load environment variables: source ./utility/environ.sh, and try to run the KAIROS agent locally (./utility/evaluation_locally.sh --verbose) to make sure everything is working properly.

If new dependencies are added to the environment.yml file, activate the environemnt and run conda env update -f environment.yml --prune.

Usage

0. Activate your Conda environment: conda activate basalt

1. Load environment variables: source ./utility/environ.sh

2. (Optional) If you modified the main KAIROS package, update it: pip install -e ./kairos_minerl/

3. Run KAIROS agent: ./utility/evaluation_locally.sh --verbose

Label Data and Retrain Agent

Instruction to label data for the state-classifier is available at https://docs.google.com/document/d/11RxGh40WVZY1RX0734E0bWiHmpVX0lJMs7-paU1dqY8/edit#heading=h.fdoz4yxhczbg.