Through this project, the goal is to get a car to learn how to self-drive on a road applying Deep Reinforcement Q-learning methods, combining Reinforcement Learning and Deep learning. The simulation will be done on the SUMO traffic simulator and with the help of TraCi a python library.
We consider a stretch of highway with a given number of lanes and we have a certain number of cars following each other on a road where passing is done by lane change. We will focus on one car which we control. Using the Deep Q-Learning algorithm, the car will have to make a decision and change or not the lane, it can go to the left, to the right or remain in the same lane.
The project aims to develop a reinforcement learning application for an agent to drive safely in dense traffic conditions. In a dense traffic situation. To do this, SUMO was used to simulate the behavior of the target vehicle and a network of autonomous vehicles and to train a model of reinforcement learning algorithms, namely DQN.
First install SUMO and then the required packages from requirements.txt
I wrote a report about this. You can find it inside report folder