Overview / Usage

Designing and Developing Reinforcement Learning Agent who controls and making decision according in the Traffic. Agent making decisions in a neural network that drives single vehicle (or multiple vehicles) as fast as possible through dense traffic. It is a game application, in which designing motion planning algorithm in order to run/drive autonomous vehicle as fast as possible in high and deep traffic.

Autonomous Agent runs the algorithm and takes the following actions to drive in dense traffic.

1. Accelerate
2. Decelerate
3. Change in to the left lane
4. Change in to the right lane
5. Do nothing ( i.e maintain speed in current lane )

Methodology / Approach

It is inspired from MIT Self Driving Cars Course.

Deep Q Learning : It takes the input from Deep Neural Networks and learning the Agent through Deep Q-Learning for taking the actions from action space (accelerate, decelerate, change to left lane, change to right lane and do nothing).

Below are technical screen shots describe how Q-Learning works, Reinforcement Learning (Deep Q-learning) and Neural Networks get communicated.

Technologies Used

AI Techniques:
Reinforcement Learning Concepts , Machine Learning, Deep Learning, Computer Vision: Convolution Neural Networks (CNNs), Deep Q-Learning, etc.,

Tools/Software:
Datasets for CNNs,
Numpy,
Python,
Keras,
TensorFlow,
Open AI GYM, etc.,

Other links

• MIT Self Driving Cars