Home

Awesome

<img src="images/packit.gif" align="right" width="30%"/>

PackIt: A Virtual Environment for Geometric Planning
Ankit Goyal, Jia Deng
International Conference on Machine Learning (ICML), 2020

@inproceedings{goyal2020packit,
  title={PackIt: A Virtual Environment for Geometric Planning},
  author={Goyal, Ankit and Deng, Jia},
  booktitle={International Conference on Machine Learning},
  year={2020},
}

For reproducing the results of the paper and using our generated dataset, use the main repository. This repository contains code for generating more data. We recommend using the main repository first and only use this one if more data is required.

Getting Started

First download/clone the repository. We would refer to the directory containing the code as <packit_dir>.

git clone git@github.com:princeton-vl/PackIt_Extra.git

Requirements

The PackIt environment is build using Unity ML-agents version 0.4.0. We provide standlone Unity builds for Linux x86_64 platform. For creating the builds, we use Unity version 2017.4.3.

Download Data

Make sure you are in <packit_dir>. download.sh script downloads all the data, unzips them and, places them at correct locations. Note that symbolic links are created between the <packit_dir>/unity-build/StreamingAssets folder and appropriate locations inside the unity data directory. This is necessary so that the data is accessible from the build unity game.

chmod +x download.sh
./download.sh

Code Organization

Evolving new packs

  1. Make sure you have downloded everything before this step.
  2. To generate a new pack, use the <packit_dir>/unity-build/generate.sh script. To generate 5 new packs for the training data first cd unity-build and the use the command ./generate.sh --dataset tr --num_packs 5 --run 201. This would generate packs at <packit_dir>/unity-build/StreamingAssets/pack_tr/201_tr. For faster loading of data, we precompute some voxel features which would be generated in <packit_dir>/unity-build/StreamingAssets/pack_tr_precompute_<unity/python>/201_tr_<0/1/2/3/4>.

Code for making you own unity build

We provide standlone unity build. In case you want to create your own unity builds follow the following steps. We strongly recommend familiarizing oneself with Unity first. Follow the steps below before making any unity build.

  1. Download and install Unity 3D. We recommend installing version 2017.4.3. Make sure you select the required Build Support while installing Unity. For example, select Linux Build Support if you want to make builds for Linux.

  2. Place the contents of <packit_dir>/unity-assets inside the Assets folder. The Assets folder should look like the following:

    Assets

  3. To set up ML-Agents inside Unity, go to Edit > Project Settings > Player. In (PC, Mac and Linux Standalone, iOS or Android), go to the Other Settings section. Then, select Scripting Runtime Version to Experimental (.NET 4.6 Equivalent) option.

  4. (Optional) You can place the <packit_dir>/unity-build/StreamingAssets folder inside Assets. This would allow running the environment inside the editor.

  5. Now follow the steps of the particular unity build you want to make: