StarkLego

Introduction

Welcome to StarkLego! You can use this python library to access different environments to interact with your RL agents, or you can even construct your own .ldr files using the built in tools that StarkLego provides.

Setting up Development Environment

The library was written in Python 3.5.6. When you install the library using pip, the library should already download all of the required libraries to run StarkLego. However, if this does not happen then you can download the following pip libraries manually:

Library	Recommended Version
tensorflow	1.8.0
gym	latest
numpy	1.16
pyldraw	0.8.2

You can also run the following line in your command-line:
pip install gym pyldraw==0.8.2 numpy==1.16.0 tensorflow==1.8.0

IMPORTANT: Create a ldraw-license.txt file in the directory wherever you are running your code from. This is necessary because in order to use the ldraw libraries you will need to include this.

Lego Piece Support

So far, only the following Lego Pieces are supported:

• 2X2 Brick Due to the lack of support for more than the 2X2 Brick, there is no customization available.

Constructing Your Own LDR Files

You might want to use this library to construct your own LDR files which contain different parts. The following is an example of how to use the StarkLego package to create a Lego World.

from StarkLego.lego_builders.service.builder import LegoWorld
from StarkLego.lego_builders.service.builder import TwoXTwoBlock

my_lego_world = LegoWorld(6,12,6)

my_lego_world.add_part_to_world(part=TwoXTwoBlock(), x=0, z=0)
my_lego_world.add_part_to_world(TwoXTwoBlock(), 2, 2)
my_lego_world.add_part_to_world(TwoXTwoBlock(), 2, 2)

print(my_lego_world.ldraw_content)
# Expected output 
#1 272 0.000000 0.000000 0.000000 1.000000 0.000000 0.000000 0.000000 1.000000 0.000000 0.000000 0.000000 1.000000 3003.DAT
#1 272 40.000000 0.000000 40.000000 1.000000 0.000000 0.000000 0.000000 1.000000 0.000000 0.000000 0.000000 1.000000 3003.DAT
#1 272 40.000000 -24.000000 40.000000 1.000000 0.000000 0.000000 0.000000 1.000000 0.000000 0.000000 0.000000 1.000000 3003.DAT

Using Environments For Training Agents

These environments cater to the agents available in the stable_baselines RL library. You can get stable_baselines by running the following in your command-line:
pip install stable-baselines==2.9.0

How to run

If you wish to run one of these environments, please feel free using the code below:

from StarkLego.environments.env_low_height import LegoEnv
from stable_baselines.common.policies import MlpPolicy
from stable_baselines.common.vec_env import DummyVecEnv
from stable_baselines import PPO2
import numpy as np


env = DummyVecEnv([lambda: LegoEnv(4, 14, 4, 4)])

model = PPO2(MlpPolicy, env, verbose=1, learning_rate=0.0001, gamma=1)
model.learn(total_timesteps=3)
obs = env.reset()

print("Done training")

for i in range(4):
    action, _states = model.predict(obs, deterministic=True)   #determinstic is `False` by default
    obs, rewards, done, info = env.step(action)
    env.render()
    if done:
        print(info[0].get("ldrContent"))   #print the state through `info` due to environment resetting

List of Supported Environments

env_low_height

The goal is to minimize the height of the Lego build.

Space	Data Type
action_space	spaces.Box
observation_space	spaces.Box

The only specifications than can be made are the dimensions of the LEGO World, and the number of pieces per build iteration.

Constructor:LegoEnv(x, y, z, noLegoPieces)

• x: the maximum x dimenstion of the Lego World
• y: the maximum y dimenstion of the Lego World
• z: the maximum z dimenstion of the Lego World
• noLegoPieces: The number of Lego pieces to be inserted into the Lego World

This environment does not allow any customization for which lego pieces can be used.