The mcenv from claucambra

mcenv's Introduction

Modified MineDojo Package

The code is a slightly modified version of Minedojo.

Installation

Please refer to this document for the package installation and use. Note that Python>=3.9 is required.

Install modified MineDojo:

pip uninstall minedojo
Clone this repo. 
Run `python setup.py install`

Modifications on MineDojo

We allow repeatedly making a programmatic task.
In minedojo.make(), we allow overriding all the default parameters in programmatic tasks (e.g. lidar_rays, initial_inventory).
We add some mirror URLs to load the simulator faster.
We support randomly teleporting the player in the fast_reset mode: when calling env.reset(), the player will be teleported to a random location near its initial spawn point.

In minedojo.make(), use fast_reset=True to enable random teleportation:

env = minedojo.make(
            task_id="harvest_wool_with_shears_and_sheep",
            image_size=(160, 256),
            fast_reset=True,
            fast_reset_random_teleport_range_low=0,
            fast_reset_random_teleport_range_high=100,
        )

Then calling env.reset(), the player will be teleported in a distance range of [0,100] by default. To disable moving the agent, use env.reset(move_flag = False).

If you don't set the fast_reset parameters in minedojo.make(), our environment is the same as the original MineDojo.

Reference

If you find the environment useful, please consider citing Minedojo.

@article{fan2022minedojo,
  title   = {MineDojo: Building Open-Ended Embodied Agents with Internet-Scale Knowledge},
  author  = {Linxi Fan and Guanzhi Wang and Yunfan Jiang and Ajay Mandlekar and Yuncong Yang and Haoyi Zhu and Andrew Tang and De-An Huang and Yuke Zhu and Anima Anandkumar},
  year    = {2022},
  journal = {arXiv preprint arXiv: Arxiv-2206.08853}
}

Recommend Projects