Deep Reinforcement Learning (PPO) car simulator using unity engine and python.

• Windows10 (64bit)
• Python 3.6
• Anaconda3 5.0.1
• Tensorflow 1.7.1

CPU: Intel(R) Core(TM) i5-4200U CPU @ 1.60GHz 2.30GHz

• python
• C#

using unity SDK ml-agents

PPO (Proximal Policy Optimization)

• RL-3D-simulator is unity project for 3D car simulator,ready to be trained.
• RL-Unity is unity project for 2D car simulator, ready to be trained.
• UnityGame is unity project for the 2D game only without ml-agents setup and scripts.

1. Download unitySDK repo https://github.com/Unity-Technologies/ml-agents
2. Install the ml-agents as shown here https://github.com/Unity-Technologies/ml-agents/blob/master/docs/Installation.md
3. download this repo,then open the RL-3D-simulator unity project in unity and goto Assets->ml-agents->myCar->scene then click SampleScene.
4. To train the game https://github.com/Unity-Technologies/ml-agents/blob/master/docs/Training-ML-Agents.md

Opponent cars position agent car position

The action of the vehicle is as follows.

• Do nothing
• move left
• move right
• move backward
• move forward

• -1 when agent collide the opponent cars
• -1 when agent collide with road boundaries
• 0.5 when agent moves forward
• 0.2 when agent moves right or left

after 2 Hours training with PPO algorithm