DDPG-CARLA

Update: This repository has been deprecated due to some hard code. https://github.com/zhangfuyang/rl_CARLA.git is the new repository for this project. If you are interested with my work, please go to this repo.

Welcome to my repo! My experiment environment is Ubuntu16.04, python3.5 and tensorflow1.12.0. If you have any issue about the code, please don't hesitate to contact me.

Installation

Clone the git repository

$ git clone https://github.com/zhangfuyang/DDPG-CARLA.git

$ cd DDPG-CARLA & pip -r requirements.txt

(Use $DDPG_DIR as the root directory of the source)

Install Carla Simulator

Download the Carla you can just download the compiled version from here . Make sure the version of the simulator is 0.8.2(stable), I'm not sure if other development or old stable versions are compatible.
extract to the directory you want $ tar -xvf CARLA_0.8.2.tar.gz $CARLA_DIR

Training

First start the Carla server cd $CARLA_DIR ./CarlaUE4.sh -carla-server -benchmark -fps=10
Run ddpg_main.py cd $DDPG_DIR python ddpg_main.py

Some Detail of implementation

Use pretrain Network https://github.com/carla-simulator/imitation-learning to extract feature. Then connected with actor and critic net.

shenhe111 / DDPG-CARLA