robot_dqn

I needed an environment to experiment with robotics and deep learning. Actually doing RL on the actual robot itself would be computational expensive. Chances are good you know that already :) So we are speaking about continious control and DQN. Truth spoken that simply isnt a good solution until you have a low dimensional action output. So its more ore less a lab solution but interesting anyway.

Getting Started

if you want to see whats going one you can draw the the robots progress with "drawRobot()" coming from "robot_arena" When you finished learning the model its really fun to do so if you execute your model.

Startig is easy

• "robot_local_learn" saves a .h5 model to your projects folder
• "robot_local_execute" runs the model you save
• "robot_server_execute" send commands to a random client e.g robot

Prerequisites

The faster your pc is the better ;)

Installing

You do need a small number of libraries.

• numpy
• keras
• pygame

Depending on your system i would recommend to do it with npm.

Acknowledgments

if youre curious its fun to start with what i've made but ML is a wide and deep complex there is so much to learn. Maybe this stuff is helpful for starting your own little project !