RL_Racer

Ros impelmnation of deep learning tecniques for robot naviagation and facial recognition

Prerequisites

Have ROS meliodic enviroment installed

Installation Guide(Facial Recognition):

pip install face_recognition

pip install opencv-python (takes ~20 minutes)

pip install numpy

chmod +x [Files name]

Installation Guide(Deep RL agents):

I recommend creating a conda environment using python 2.7 for this project.

Follow directions at https://www.anaconda.com/distribution/#download-section to install anaconda

Then:

conda install numpy

conda install pytorch torchvision cudatoolkit=10.1 -c pytorch (change cudatoolkit version as necessary)

conda install matplotlib

Running Guide(Facial Recognition):

1. be in outermost file

2. run bu shortcut on a robot

3. run roslaunch turtlebot3_bringup turtlebot3_rpicamera.launch framerate:=20

4. rqt_image_view and select the compressed file you should see this

5. While in RL_racer folder type into terminal $rosrun RL_racer face_finder.py

Running Guide(Deep RL agents)

To launch the gazebo simulation of small_track1 do:

roslaunch turtlebot3_gazebo small_track1.launch

1. Launching a Deep Deterministic Policy Gradient agent (DDPG)

roslaunch deepRL train_ddpg_agent_multistart.launch

2. Launching a Soft Actor-Critic agent (SAC)

roslaunch deepRL train_sac_agent_multistart.launch

3. Drive robot using teleop and store experience

roslaunch deepRL make_experience.launch

4. Launch an agent and see how it performs without noise

roslaunch deepRL test_agent.launch