lelmac / robotsim

Simulator for EV3 as Environment for OpenAI Gym

4 Environments:

• AutonomousRobot-v0: Discrete Action Space, Infraredsensor and Ultrasonic Sensor, Random Spawns
• AutonomousRobot-v1: Continuous Action Space, Infraredsensor and Ultrasonic Sensor, Random Spawns
• AutonomousRobot-v2: Discrete Observation Space, Discrete Action Space (for Q-Learning), Infrared & Ultrasonic Sensor
• AutonomousRobot-v3: Discrete Action Space, Target Position, Random Spawns, 3 Distance Sensors (Basically Scenario 3 from the Thesis)

Each Environment can of course be customized indivually.

Setup

1. Clone the Repo
2. Install OpenAI Gym: https://github.com/openai/gym
3. Run:

  sudo pip install -r requirements
  sudo python setup.py install

OR: Use the Dockerfile to create a Container

Then you should be ready to use the Environments by import gym_robot:

  import gym_robot

Misc

The Repository comes with a few extas:

• Few Agents: DDPG, DQN, DQN (With Keras-RL), Q-Learning, A3C(not really threaded), Keyboard Agent (can be handy when designing environments)

• A Dockerfile

• 2 Scripts for visualizing logfiles:

  • for keras-rl logs:

  visualize.py filename.ending
  

  • for csv files:

  cs.py filename.ending 
  

• Some pretrained Models

Simulation:

• The simulation is located in gym_robot/envs/obstacles.py
• Got 2 classes: one for obstacles and one for the robot itself
• The Methods for the sensors are in robot class
• Rendering is done by OpenAI Gym Renderer in the specific environment
• test.py contains unit tests for the classes